LINK MAINTENANCE AND CHANNEL EVALUATION TECHNIQUES FOR HF RADIOCOMMUNICATION LINKS

Transcription

1 LINK MAINTENANCE AND CHANNEL EVALUATION TECHNIQUES FOR HF RADIOCOMMUNICATION LINKS Thesis submitted for the degree of Doctor of Philosophy at the University o f Leicester by David Jonathan Brown B.Eng. M.Sc. Department of Engineering University o f Leicester January 2001

2 UMI Number: U All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion. Dissertation Publishing UMI U Published by ProQuest LLC Copyright in the Dissertation held by the Author. Microform Edition ProQuest LLC. All rights reserved. This work is protected against unauthorized copying under Title 17, United States Code. ProQuest LLC 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, Ml

3 Link Maintenance and Channel Evaluation Techniques for HF Radiocommunication Links. David Jonathan Brown. Abstract Methods for estimating the prevailing channel characteristics for use in an automatic modem waveform selection management system for high frequency (HF) radio circuits are addressed in this thesis. Current techniques for channel evaluation and management are reviewed in detail and their limitations noted. Various channel probe signals are considered including Barker sequences, long sequences found by means of neural network methods, modified Legendre sequences and a sequence which forms part of the STANAG-4285 waveform. Experiments have been undertaken in which the new methods of channel probing are evaluated and compared with those currently in use. A prototype system that automatically changes the modem data rate based on measured frame error rate is demonstrated. This system can significantly improve data throughputs compared with fixed data rate systems. The use of in-built or add-on channel probing methods that do not require breaks in the data transmission are promoted over dedicated probing methods which require a break in the normal data communication signal. The concept of reducing the relative amplitude of an add-on probe (and hence improve the SNIR) by using a signal cancelling technique at the receiver is introduced. Finally, recommendations for additional work necessary to realise an automated modem waveform management system to complement existing frequency management systems are made.

4 Acknowledgements I am grateful to the staff of the Department of Engineering and to the members of the Radio Systems and the Radio and Space Plasma Physics research groups. In particular, I am indebted to Dr. Mike Warrington and Dr. Alan Stocker for their endless help and encouragement provided over the time it has taken to complete this work. The Defence Evaluation and Research Agency funded this work to provide an academic insight into the problems facing modem digital HF communications. This allowed me to carry out interesting and rewarding research during the three years of my postgraduate study at Leicester. Subsequently DERA employed me to carry on this study and to take up similarly engrossing work. I would like to acknowledge all the help, encouragement and financial support that DERA has provided and I would particularly wish to thank Dr. Andy Gillespie and Mrs. Sue Trinder for their continued commitment to the future of modem digital HF communications and for their input to my ongoing work. Whilst my three years at Leicester were thoroughly enjoyable and I was driven by the research work, I did rely on the financial and moral support of my mum, Angela, for which I am eternally grateful. My extended period of writing this thesis whilst learning yet more, in my new role as DERA employee, would have been impossible if it were not for the tireless support of my Fiancee, Rachael who also kept me motivated by feeding me as I spent countless hours in from of, and inside, the computer.

5 Contents CHAPTER 1 - INTRODUCTION T h e IONOSPHERE Ionospheric regions H ig h f r e q u e n c y r a d io w a v e p r o p a g a t io n Sky wave propagation Ground wave propagation Multi-moded propagation Propagation prediction D a t a c o m m u n ic a t io n s o v e r H F c h a n n e l s FSK modulation schemes PSK modulation schemes QAM modulation schemes W o r k in g w it h pr o pa g a t io n p r o b l e m s Channel parameters Channel assessment Overcoming propagation effects Automated HF radio communications T h e p r e s e n t in v e s t ig a t io n CHAPTER 2 - A REVIEW OF LINK MAINTENANCE AND CHANNEL EVALUATION TECHNIQUES IONOSONDES The vertical incidence ionosonde The oblique incidence ionosonde Using vertical ionograms to estimate oblique paths Ionosonde waveforms D e d ic a t e d io n o s p h e r ic p r o b e s Coded pulses Limitations o f coded pulses D e r iv e d c h a n n e l s o u n d in g U s in g c h a n n e l s o u n d in g s f o r l in k e s t a b l is h m e n t a n d m a in t e n a n c e C o n c l u d in g r e m a r k s CHAPTER 3 - EXPERIMENTAL CONFIGURATION In t r o d u c t io n Common system structure Location...3-2

6 T h e t r a n s m it t e r Hardware Software T h e RECEIVER Hardware Software V a l id a t io n o f e q u ip m e n t CHAPTER 4 - COMPARISONS OF DIFFERENT MODULATIONS OVER THE COURSE OF SEVERAL DAYS E x p e r im e n t a l c o n f ig u r a t io n C o m p a r is o n s o f m o d u l a t io n t y p e s Data collected on 26/4/ Data collected on 27/4/ Data collected on 13/6/ Data collected on 28/6/ Data collected on 29/6/ Data collected on 30/6/ C o n c l u d in g r e m a r k s CHAPTER 5 - ERROR RATE DERIVED LINK MANAGEMENT Il l u s t r a t iv e pe r f o r m a n c e d a t a f o r v a r io u s m o d u l a t io n s c h e m e s H o w ERROR RATE DERIVED MANAGEMENT COULD BE ACHIEVED DERA data rate change test bed Results o f error rate derived management experiments C o n c l u d in g r e m a r k s CHAPTER 6 - IONOSPHERIC PROBING TO AID MODEM WAVEFORM SELECTION D a t a m o d u l a t io n pe r fo r m a n c e c h a r a c t e r is t ic s Multi-dimensional characterisation o f modem performance Ionospheric conditions to be expected Modulation types available for differing channel parameters T h e n e e d t o m e a s u r e o r e s t im a t e c h a n n e l p a r a m e t e r s t o a id m o d u l a t io n -t y p e s e l e c t io n 6-6 U sin g th e B a r k e r - 13 se q u e n c e a s a p r o b e Results from on-air Bar her-13 probe experiments S T A N A G EQUALISER TRAINING SEQUENCE Example scatter gram Limitations o f using the STANAG-4285 synchronisation sequence fo r channel evaluation E x a m p l e s o f c h a n n e l e v a l u a t io n p a r a m e t e r s a n d m e a s u r e d m o d e m p e r f o r m a n c e o v e r th e A n g l e - L e ic e s t e r p a t h C o n c l u d in g r e m a r k s

7 CHAPTER 7 - CHANNEL PROBING BY THE ADDITION OF A PULSE COMPRESSION CODE TO THE MODEM WAVEFORM T y p e s o f pr o b in g s e q u e n c e s A 229 BIT MODIFIED LEGENDRE SEQUENCE FOR CHANNEL EVALUATION Limitations and theoretical performance o f code bit modified Legendre sequence probe format Experimental configuration Results o f probing with 229-bit modified Legendre sequence The effects o f the 229 bit probe waveform on the performance o f a data modulation E x c is io n o f a d d -o n c h a n n e l e v a l u a t io n pr o b e w a v e f o r m C o n c l u d in g r e m a r k s CHAPTER 8 - CONCLUSIONS AND RECOMMENDATIONS FOR FURTHER WORK A n a u t o m a t e d H F c o m m u n ic a t io n s s y s t e m REFERENCES...I

8 Chapter 1 - Introduction The ionosphere Ionospheric regions Surrounding the earth between altitudes of around 90 and 600 km is a highly ionised layer of the atmosphere. This layer is one of several atmospheric layers, each having varying attributes. Some of these layers are identified in terms of their temperature, and some in terms of the state of their ionisation. It is this ionised layer that is of greatest interest to long distance communicators as it makes beyond line of sight, high frequency, radio communications possible. The effects of solar radiation on the atmospheric gases that are present around the earth cause ionisation. Here, the ultraviolet light from the sun strips electrons from neutral atmospheric atoms thus making them positively charged. These atoms, which are then known as ions due to their charge, and more importantly the liberated electrons, form the ionised and hence conductive layer of the atmosphere which is called the ionosphere. The concentration of free electrons is dependent on solar radiation and as such is subject to diurnal, seasonal, 11 year solar cycle and geographical variations. The electron density also varies with height owing to the density of atoms which can be ionised decreasing at higher altitudes and to the ultraviolet light intensity decreasing at lower altitudes. Figure 1.1 illustrates two sample electron density curves as functions of height for daytime and for night-time conditions. Figure 1.2 shows that the electron density differs diumally and with geographical location. For the same geographic locations, Figure 1.3 demonstrates seasonal changes in the profiles too. 1-1

9 h- X O UJ X ELECTRON DENSITY (ELECTRONS/m3 ) Figure Sample electron density profiles ((a) Daytime, (b) Nighttime) (After Davies 1990). 1-2

10 AVERAGE QUIET-DAY PROFILES MARCH 1959 r IV "T' T I \ I a 1' I ' V IT - H " \ \ \ \ \ \ NEWFOUNDLAND " \ \ \ \ \ \ ss V \ N ' ' \ \ ' ~ y y \ \ \ N r ) \\ " " \ \ il... I 'v " 11 v " ' V " I " ' IV... -\ '\ '\ 1 \ 1 \ 1 \ 1 \ 1\ ' I ' N N S ' \ \ \ \ X \ \ \ - x 400 o x GRAND BAHAMA f I x ELECTRONS/m * HUANCAYO - (MARCH 1960)- Figure Examples o f electron density as a function o f height, location and time o f day (After Davies 1990). 800 V T,_r \ \ 400 zoo o 800 E 600 o 400 zoo MIDNIGHT MEAN N(h) PROFILES iv" " it...r 1 \ _ \ ' \ \ * A /> / ) may 59 MAY ' '59 JULY SEPT NOV 1 '59 1 '59 1 *59 JULY '59 \ \ SEPT '59 NOV '59 r "» t NEWFOUNDLAND i i \ i / / : JAN MAR 1 ' ' JAN '60 TV GRAND BAHAMA MAR '59 HUANCAYO - ZOO 0 MAY JULY SEPT NOV '60 I '60 I '60 I '60 X 1011 ELECTRONS/m3 Figure Examples o f electron density as a function o f height, location and season (After Davies 1990). 1-3

11 The peaks seen in the electron density profiles are associated with individual regions within the ionosphere and are known as the D, E and F regions, see Figure 1.4. The E region was the first of these regions to be discovered, with E denoting electric field. The E region exists from 95 to 120 km, under certain conditions a thin layer of this region becomes exceptionally ionised and this is then known as sporadic-e and is denoted in text as Es. The D region exists from 60 to 95 km, at night the electrons here recombine with the ions at a greater rate than during the day as the production of ions is lower, and hence this region effectively disappears. The highest region is the F layer, which extends from 120 km upwards. During the summer months this region may split into two - the FI and F2 region, the FI region mostly disappears at night. DAYTIME IO N O SPH ERE NIGHTTIME IO N O SPH E R E Figure Illustration o f ionospheric regions during the day and during the night (After McNamara 1991). 1-4

12 High frequency radio wave propagation Sky wave propagation Radio propagation in the high frequency (HF) bands can facilitate communications over long distances by means of sky wave paths. Sky wave propagation relies on the radio waves being transmitted and effectively reflected back towards the earth s surface from the ionosphere to the receiver. There can be one or more reflections from the ionosphere together with intermediate ground reflections, plus complicated ducting effects between individual layers of the ionosphere (Figure 1.5). Though it is common to treat the effects described here of the ionosphere as reflections, they are in fact refractions of the radio waves. Flayer Flayer 2F F layer F layer Elayer F-E E layer EF Figure 1.5 Schematic illustration o f several different propagation modes. Ground wave propagation Propagation of HF signals is also supported in the absence of reflections from the ionosphere and is called ground wave propagation. Here signals travel from the transmitting antenna to the receiving antenna via a path that follows the curvature of the earth. The distance achievable via ground wave is dependant on ground conductivity and also the frequency used (at 10 MHz, ground wave will typically be useful to about 60 km). The use of ground wave is of less interest than skywave due to the shorter paths possible, though this may be advantageous for some types of communication. 1-5

13 Multi-moded propagation The ionosphere does not allow unlimited world-wide communications. The radio waves that travel between a transmitter and a receiver can be viewed as rays. In a very simple scenario the ray path starts from the transmitter, goes up into the ionosphere is reflected at the mid-point and then propagates back down to the receiver. However, in practice there are often at least two ray paths of differing lengths - hence the signal can arrive at the receiver via two or more paths (Figure 1.6) with differing time delays, this is known as multi-moded propagation. Multi-moded propagation can result in a phenomenon known as fading, this can cause severe communications difficulties owing to the effect it has on the received signal. Fading will be discussed later in this chapter along with some other effects that are due to the changing nature of the ionosphere. Flayer Transmitter Receiver Figure 1.6 Example o f two moded propagation. Propagation prediction As mentioned earlier, the regions that reflect HF radio waves are a changing natural phenomenon. The presence and structure of the reflective regions have been studied for many years and these studies have resulted in the availability of propagation prediction methods. Such methods enable users to predict to a certain degree of accuracy the likelihood of being able to communicate between two stations on a given frequency at a given time. These are often used for link planning. 1-6

14 Long term forecasting techniques predict monthly propagation parameters based on historical ionospheric data. From these, parameters such as the optimum working frequencies for given days can be extracted. These predictions do not take into account the effects of ionospheric storms, sporadic-e and interference from other users, all of which have a bearing on the ability to communicate. One widely used prediction package is the Ionospheric Communications Analysis and Prediction (IONCAP) computer program, this is described by Teters et al [1983]. Given details of the transmit and receive antennas, locations, date, time and sunspot number, IONCAP can produce predictions of estimated SNR for varying frequencies amongst other reports. Short term forecasting is based on observations of the sun and of frequencies being supported on a given circuit. It is used to increase communications reliability by improving the long-term models based on observations made in the recent past. However, this method still does not necessarily indicate which of the available frequencies is the best to use. Nowcasting is a term applied to predictions of the behaviour on HF paths based on the current solar and geomagnetic activities. These figures are used to calibrate the results obtained from ionospheric monitoring systems sited at sub-optimal locations in order to extrapolate to the locations of interest. Ideally, monitoring, or probing of the ionosphere would take place over the same path as desired for communications but this is often not possible due to the specialised nature of the equipment required. Data communications over HF channels FSK modulation schemes HF data communications methods have often adopted FSK modulation, this is simple to implement and is widely used. With FSK, a sub-carrier frequency within the channel bandwidth is shifted to one of a number of discrete frequencies to signal transmit information (i.e. each transmit symbol is encoded as one of a number of possible frequencies). Binary FSK In the case of binary FSK, only two transmit symbols are needed and these map directly to two discrete frequencies within the channel bandwidth. This type of modulation scheme is affected by time dispersion and Doppler shifts. Time dispersion causes receive errors 1-7

15 because of inter-symbol interference. Inter-symbol interference (ISI) is a problem associated with communication systems that experience multi-path effects, that is where two copies of the transmitted signal arrive at the receiver with a time delay (At) between them. Typical values of At for HF paths range up to around 5 ms, at this figure, data modulations such as binary FSK will suffer if their rate exceeds around 75 baud, this is because a time delayed symbol may interfere excessively with a subsequent, non-timedelayed symbol (i.e. the two may partially overlap). Doppler shifts however, will only affect FSK signals if the frequency shift of the keying is less than the Doppler shift caused by the propagation effects. To make the waveforms more robust in times of difficult propagation conditions the signalling rate has historically been reduced and the frequency shift of the modulation has been increased. Although this can be effective in reducing the bit error rate, these techniques also reduce the data throughput and increase the bandwidth occupied therefore making less efficient use of the already crowded HF spectrum. Today, more sophisticated methods are being developed and deployed to allow data to be transmitted faster and more reliably using the spectrum as efficiently possible. Multiple frequency shift keying A modulation method known as multiple frequency shift keying (MFSK) has been used to increase robustness of data communications systems to ionospheric effects. With MFSK, multiple frequency pairs are used to each convey a number of bits of data, the advantage here is that the signalling rate is lower than if a single frequency pair were to be used (as with regular FSK), hence the likelihood of ISI is reduced. Unfortunately, even across the channel bandwidth, frequency selective fading can occur and cause the loss of one or more tones and hence some disruption to the transmission. Honary et al [1992] discuss one MFSK technique and a method of forward error correction aimed at increasing the robustness of HF communications. PSK modulation schemes An alternative method of transmitting data symbols across a communications bearer is to shift the phase of a sinusoidal carrier. Here the unmodulated carrier takes the form of a discrete frequency within the channel bandwidth and has its phase periodically shifted by various quantities to signal different symbols. 1-8

16 ST AN AG-4285 In this NATO (North Atlantic Treaty Organisation) standard, [NATO, 1989], a 1200/2400/3600 bps (bits per second) single tone modem for HF radio links is characterised. This employs a robust modulation scheme based on phase modulating a 1800 Hz sub-carrier. One of eight phase values may be selected (8-ary PSK) at a fixed rate of 2400 baud. A process called transcoding is used to link a sequence of up to three bits of data with one transmit symbol, this allows a higher data rate than the modulation keying rate. The constellation diagram in Figure 1.7 demonstrates three bits of data being transcoded to one transmit symbol. With three data bits being encoded to one transmit symbol at the rate of 2400 baud, a raw data rate of 7200 bps can be achieved, though this is reduced to a maximum of 3600 bps in STANAG-4285 due to the use of half of the transmit symbols for synchronisation and channel equalisation purposes. The constellation diagram shows that only one bit of the three changes between adjacent phases, this minimises the number of errors caused should an adjacent phase be detected in place of the true phase. Channel equalisation is employed when demodulating the STANAG-4285 waveform, this uses derived channel parameters to reduce the effects of HF propagation characteristics imposed on a signal. Two methods of channel equalisation are described in the STANAG document. A modulation scheme such as STANAG-4285 has differing channel requirements from traditional FSK modulations. Time dispersion is less of a problem with STANAG-4285 as the channel equaliser, as mentioned above, is intended to cope with delay spreads up to around 7 ms. Frequency dispersion is a limiting factor for PSK modulation and measurement of the channel scattering function would be a useful feature of a suitable link establishment and maintenance system. 1-9

17 0 degrees, symbol 0, data degrees, symbol 7, data degrees, symbol 1, data degrees, symbol 6, data degrees, symbol 2, data degrees, symbol 5, data degrees, symbol 3, data degrees, symbol 4, data 111 Figure Phase constellation diagram showing encoding o f data bits on to transmit symbols for an 8-ary PSK waveform. MIL-STD A The waveform described in the US Military standard MIL-STD A [US DoD, 1991] is similar to STANAG-4285 as both use 8-ary PSK and a frame based approach to carrying their payload data. Forward error correction and receiver equalisation are employed to improve performance. Data rates supported by this scheme range from 75 to 4800 bps. QAM modulation schemes By modulating both the phase and the amplitude of a sinusoidal carrier, it is possible to use an even greater number of transmit symbols and hence potentially increase the user data rate further than achieved using the techniques described so far. This modulation method is termed quadrature amplitude modulation (QAM). MIL-S TD B Although the STANAG-4285 and MIL-STD A waveforms can support up to 3600 bps and 4800 bps respectively in a standard 3 khz HF channel, a newer standard has been developed and this describes a family of waveforms that can achieve bps in the same bandwidth. This standard, MIL-STD B [US DoD, 2000] features data rates of 3200, 4800, 6400, 8000, 9600 and bps with varying interleaver lengths. The data rates from 6400 bps and upwards use QAM waveforms. Due to advances in signal 1-10

18 processing technology and computing power, the lowest two data rates of this standard easily outperform the highest two data rates of STANAG-4285 in terms of bit error rate for a given SNR. However, it should be noted that the higher data rates (6400 bps or greater) of this new standard require channel conditions far in excess of those widely used for 2400 bps PSK. NATO will shortly adopt the waveforms of MIL-STD B as STANAG Working with propagation problems Channel parameters In order to determine the most appropriate modulation format to transmit data over a given HF channel, it is necessary to have some knowledge of that channel s current state. Or, in other words, some parameters relating to the performance of the channel need to be evaluated prior to selecting the modulation technique to use for communicating. Darnell [1978] defines real time channel evaluation as: Real Time Channel Evaluation is the term used to describe the process o f measuring appropriate parameters of a set o f communications channels in real time and employing the data thus obtained to describe quantitatively the states o f those channels and hence the capabilities for passing a given class, or classes, o f communications traffic These parameters can change with ionospheric conditions, with other HF user s signals and with any motion or changes associated with either the transmitter or receiver stations. As discussed earlier in this chapter, sky wave propagation is variable and can result in several ray paths between the transmitter and receiver. This has the effect that multiple, time delayed copies of a signal may be detected at the receiver. The parameter associated with the delay between receiving the first and last significant modes is known as delay spread. Movement of the reflecting ionospheric layers or of the transmitter or receiver antennas can cause variations in the path length and consequently Doppler shifts are imposed on the signal frequency. If the path length becomes shorter during a transmission, the carrier frequency of the signal at the receiver appears to rise, the converse is true if the path length increases. The range of frequency shifts that are being experienced over a channel is known as the channel s Doppler spread characteristic. 1-11

19 Changes in signal level, or fading, is another characteristic of the HF sky wave channel. Interference fading is due to changing lengths of signal paths between the transmitter and receiver (and hence relative phases of the modes). This type of fading may be either constructive or destructive, in the former case the signals are in phase and sum together, in the latter case the signals are out of phase and tend to cancel each other out. Absorption in the D region can also cause fading effects as the signal is no longer effectively reflected back from the ionosphere. Communication may become difficult if the fading rate is fast, or if the depth of each fade, is too great. Often, ionospheric conditions will cause a change of signal characteristics on some of the available channels and may allow enhanced communications for short periods of time. This can frequently be seen at times when sporadic-e propagation exists. Sporadic-E propagation cannot be predicted reliably, but can be detected as an improved channel using channel evaluation techniques. Operating on channels that are being supported by sporadic-e maybe advantageous to the communicator if they can be identified quickly enough. Other users of the HF spectrum will inherently cause some interference due to the nature of the communications medium. This interference may be unintended, or at times may be intentional with a view to preventing communications from taking place. There are many categories of interferers, some of which may be tolerated with a small reduction in communications quality, whereas others will render a channel useless. Knowledge of the type of any interferer can be useful in channel assessments. Interference problems may be reduced in several ways depending on the nature of the interference. With co-channel interference it may be possible to filter out the unwanted signal component. Wide band interference may often be tolerated. In-channel interference is often of most concern since filtering would also affect the wanted signal. Noise sources such as man made electrical signals and background atmospheric noise can cause the proportion of useful signal energy captured by a receiver to be reduced (i.e. the SNR becomes lower). This will, if the SNR becomes sufficiently small, prevent demodulation as the modem will fail to correctly recognise and interpret the transmitted signal. 1-12

20 Channel assessment To enable reliable and efficient long distance communications using HF, the effects of ionospheric propagation need to be assessed and decisions made on the frequencies and modulations to be used. Techniques such as checking the readability of specific modulations on available channels or assessing larger sections of the HF band using a specialised sounder are often used, narrow and wide band channel evaluation methods are discussed in detail in Chapter 2. The effects of ionospheric propagation are constantly changing and any assessment needs to be updated regularly in order to maintain an accurate picture of the current state of available frequencies. A limited number of channels may be assigned to a user, none of which will be perfect, so the ability to compromise and to adapt to the prevailing conditions is important in a good communications system. Overcoming propagation effects Modem modulation schemes often mandate equaliser techniques to overcome some of the effects imposed by the ionospheric propagation environment. Errors are frequently minimised by the use of interleavers and forward error correctors that are commonplace in modem standards. Other schemes to overcome difficult conditions include having a range of waveforms and data rates which can be selected for use, i.e. a slow, but robust system can pass data even in very hostile propagation conditions, yet a fast system can take advantage of short term propagation improvements to pass data more quickly. Automated HF radio communications A collection of techniques is brought together under the title of the Automatic Radio Control System, or ARCS and is intended to simplify HF communications. Arthur and Maundrell [1996] describe the ARCS concept in detail, concluding that ARCS will benefit HF communicators by reducing workload and increasing system performance. ARCS has three components: Automatic Channel Selection (ACS); Automatic Link Establishment (ALE); Automatic Link Maintenance (ALM). 1-13

21 ACS uses prediction, forecasting and nowcasting techniques along with ionospheric channel evaluation methods to select the frequencies of interest to the communicator. Channel evaluation can show what kind of interference is present on each channel and whether it will prevent effective communications, it will also determine relevant propagation characteristics such as time and frequency dispersion. Various methods for channel evaluation are discussed in Chapter 2. The ALE system is responsible for establishing a data path between the two ends of a radio link, negotiating any encryption, setting the initial modulation, and for ensuring synchronisation of data transfer. ALM continuously monitors the link and provides the decisions on whether to change to another channel or modulation scheme. The modem waveform type may be changed to make best use of the underlying channel characteristics. In the shorter term, the data rate and other waveform specific parameters may be altered to take advantage of improved propagation or to cope with changes in SNR. In order to make efficient use of the HF channels that a communicator has available, some form of link management must be employed. Traditionally, due to the lack of automated procedures, link management has been undertaken manually with communications parameters being chosen by to what is known to work for the majority of the time, possibly aided by propagation predictions and / or simple channel evaluation. If the performance characteristics of the link can be readily derived, it may be possible to make better choices. Additionally, if these performance characteristics can be determined automatically, and the communications equipment can be controlled by computer, it is possible to automatically manage links. Link performance characteristics include any error statistics that are available for the data communications protocols that are being used over a link. Modem packet or frame based communications protocols (data link layer, or even above) used over HF will very often keep track of blocks of data that become errored during transmission, with a view to retransmitting the data as necessary. It is typical for a communications system to run with several layers of protocols being used, each with its own particular and specialised purpose. Controlling and interfacing with the modem will commonly be achieved by a data link layer protocol. One example of a data link layer protocol specifically designed for use over HF is defined in STANAG

22 [NATO, 1999]. STANAG-5066 is often used in a reliable delivery mode and will resend frames of data that fail a checksum test at the receiver, it uses an ARQ (automatic repeat request) mechanism to achieve this. At higher levels, that is on top of the data link layer protocol, network and transport protocols such as the Internet Protocol (IP) and the Transmission Control Protocol (TCP) (described in RFC 791 [Postel, 1981] and RFC 761 [Postel, 1980] respectively) are often used. The TCP layer guarantees reliable error-free delivery of data frames and implements a method of retransmitting frames based on nonreception of acknowledgements. It is possible to interface with either the STANAG-5066 or the TCP protocol layers, and in other cases similar protocols, in order to gather error statistics which relate to the performance of the underlying HF link. An error-rate parameter such as this can be used to make decisions about whether the current communications parameters are optimal for the current channel. Modem HF communications links will typically employ multiple-waveform modems partially due to the small marginal cost of including the various standard waveforms in one package. The waveforms are implemented as computer programs, with the modem hardware being made up of at least one microprocessor, some storage memory and the audio interface components. Such modems are invariably controllable via a serial data port to facilitate communications parameter programming, so switching between waveforms, data rates and forward error correction coding schemes is simple and can be readily automated. If the performance of the available waveforms is known, it should be possible to change from one waveform to another as the quality of a link changes. Consequently, it could be possible to try to maintain an error rate or data throughput rate over a link by altering the communications parameters based on the current measured error rate. Doing so could allow the link to be used to pass more data thereby allowing other resources to be freed up resulting in a cost saving (e.g., using satellite communications is typically more expensive than using HF), or allowing them to be used for other purposes. Alternatively, the same quantity of data could be passed in a shorter period of time, which is desirable in order to reduce the transmit time thus lowering the probability of detection, or to take advantage of enhanced propagation conditions which are transient in nature. Otherwise, data could be sent over the link with a uniform error rate to maximise the performance of forward error correction protocols and reduce retransmissions or the amount of errored data. 1-15

23 To allow link management to be effective, an understanding of the available waveforms is necessary. As indicated previously, the broad range of HF waveforms available have differing performance characteristics, with some being more suited to certain applications than others are. The present investigation As propagation prediction methods do not provide sufficient accuracy to allow a single HF channel and modulation scheme to be selected automatically, other methods, such as channel evaluation must be used in conjunction to enable automatic channel selection and link maintenance to take place. Real time channel evaluation (RTCE) is the title given to methods of measuring current channel characteristics, with these characteristics then being used to determine channel availability and suitability for communications [CCIR Report 889]. This work is a study of some channel evaluation techniques and how they can be used to enhance current HF data communications. A review of existing techniques has been undertaken and gives a background to the field of channel evaluation, this is presented in Chapter 2 along with a justification for the further research that has been carried out. 1-16

24 Chapter 2 - A review of link maintenance and channel evaluation techniques There have been many channel evaluation methods proposed and adopted over the last few years. This study will divide channel evaluation methods that are considered interesting and appropriate for modem digital HF communications into three categories. These techniques can gather channel parameter information in several ways including: Probing the ionosphere using ionosondes that transmit probe signals across broad sections of the HF band to measure propagation characteristics; Transmitting dedicated narrow band sounding probes in place of a data signal and analysing these probe signals to measure the channel parameters; Deriving channel parameters from the normally transmitted data signal itself. Ionosondes A system known as an ionosonde is an ionospheric sounder, and is made up of separate transmitter and receiver systems which are time and frequency synchronised [McNamara 1991]. The ionosonde works in the following manner: the transmitter radiates a known signal whilst sweeping the transmit frequency over the range of its operation, after a short delay the receiver demodulates the reflected signal and makes a comparison between sent and received signals to measure, typically, the signal amplitude and presence across the range of operation. Ionosondes typically operate over the 2-30 MHz or 2-16 MHz ranges and sweep these ranges at 100 khz/second or 50kHz/second respectively thus taking 4 minutes and 40 seconds to produce an ionogram. The advantages of using an ionosonde system for channel evaluation include their ability to detect some short term effects such as sporadic E-layer propagation and that they measure the maximum usable frequency. Multi-moded propagation and the modal structures are clearly shown in ionograms and knowledge of these can further aid the choice of channel or modulation scheme. The ionosonde is not the most ideal channel evaluation system for several reasons. An ionosonde typically transmits over the entire HF band without regard for other users, although as the time spent on a single spot frequency is negligible, any interference should 2-1

25 be minimal. The time taken to produce an ionogram is quite long and this means that the propagation data can be outdated by the time the plot is complete. It is interesting to note that Bartlett et al [1996] put forward the idea of a spectrally efficient oblique incidence ionosonde like system. This segments the HF band and uses just 5% of the spectrum that a traditional continuous sounder would to produce similar results. It is not possible to detect channels that are suffering from interference sources by using ionograms. Furthermore it is not practical due to some of the above reasons to install an ionosonde system for every HF circuit that needs channel evaluation capabilities. The vertical incidence ionosonde If the transmitter and receiver ionosonde components are co-located, the result is a vertical incidence ionogram due to the ray path from transmitter to receiver being predominately up and back down. For more than 50 years, these ionosondes have been used world-wide to study and quantify the state of the ionosphere. Two examples of vertical incidence ionograms are shown in Figure 2.1, these ionograms covers the frequency range from just below 2 MHz to over 9 MHz. The main features of the first ionogram in this figure are the E, FI and F2 mode as highlighted. The E mode can be seen to extend from around 1.6 MHz to 3.7 MHz with a small gap around 2.1 MHz. The FI mode supports propagation from 3.9 MHz to 4.75 MHz and the F2 mode from 5 MHz to 6.5 MHz. There is no evidence of multimoded propagation as there are no points where two or more modes exist at the same frequency. The second ionogram in Figure 2.1 shows clear evidence of multimoded propagation via 1 and 2 hop F modes, additionally it shows the 1 hop F mode splitting into ordinary and extraordinary components at the higher frequencies. As vertical ionosondes measure the time delay between transmission and reception at a single point, their use for link planning might seem negligible. It is however, possible to use vertical incidence data to estimate propagation conditions over a long distance path as will be described later in this chapter. 2-2

26 Boulder, Colo. 29 Jun UT ^Qp Frequency (MHz) (a) Boulder, Colo. 3 0 Jun UT fqp jx p O* <D X o Frequency (MHz) Cb) (a) (b) Sample summer daytime ionogram with ordinary wave traces only. Reflections are from the E layer, the F, layer, and the F2 layer at progressively increasing heights and frequencies. Sample ionogram taken on a summer evening in middle latitudes, showing one-hop (IF) and two-hop (2F) echoes from the F layer. The signals are split into ordinary waves (O) and extraordinary waves (X) having critical (penetration) frequencies fdf and fxf, respectively. The increase of the virtual height at the lower frequency end of the extraordinary trace is caused by retardation in underlying ionization. Figure Sample summer daytime ionograms (After Davies 1990). The oblique incidence ionosonde If the transmitter and receiver are located at different sites, as a pair o f communications stations would normally be, the result of this type o f sounding is known as an oblique 2-3

27 ionogram due to take-off angle a ray is likely to take in order to reach the receiver. It is common for the military to use oblique incidence ionosonde receivers onboard mobile platforms, such as ships, in conjunction with fixed (possibly shore) based ionosonde transmitters to determine the best frequency for long distance communications. Oblique incidence ionograms provide direct observations of all propagation modes and show the MUF for each of them. Figure 2.2 is an simulated idealised example of an oblique incidence ionogram, it indicates the traces that might be expected for 1 and 2 hop E, FI and F2 modes. The y-axis on this plot is calibrated in kilometres indicating the apparent path length from transmitter to receiver, it is also common to see this axis specified in terms of time delay. If an ionogram such as this were recorded from a real system, it would indicate at 4.5 MHz for example, three modes of propagation, 1 hop E, 2 hop E and 2 hop FI. 19 JUNE 03 UT T«R>WO 2F2 GROUP PATH 2F1 2E 1F1 1F2 «s? * 10 H IX IX M f (MHz) Figure 2.2 Simulated oblique ionogram (After McNamara 1991). Using vertical ionograms to estimate oblique paths As mentioned briefly above, it is possible to use the results from a vertical sounding to estimate the conditions over oblique paths. For this to be possible, the vertical ionosonde must be at, or very near to the mid-point of the oblique path. Such estimations are of use in situations where the use of a dedicated oblique ionosonde is not practical. 2-4

28 Davies [1990] describes equivalence theorems that show the relationship between vertical and oblique ionograms. From this, Martyn s equivalent path theorem states: Iff, and f v are equivalent frequencies of waves reflected obliquely and vertically from the same real height in a flat ionosphere, then the virtual height o f reflection o f f v is equal to the height o f the equivalent triangular path for the oblique signal. A mapping of oblique and equivalent vertical frequencies for virtual heights is required to estimate an oblique incidence ionogram from a vertical incidence ionogram. Graphically presented, this mapping is known as a transmission curve. An example set of transmission curves for a distance of 2000 km and for five different frequencies is presented in Figure 2.3 as an overlay to a vertical ionogram. The intersections of a transmission curve with the ionogram traces give the virtual heights of reflection for the transmitted signal, thus indicating propagation modes uj f. MHz Figure Family o f transmission curves for fixed 2000 km path for a flat earth and flat ionosphere (After Davies 1990). Two examples of the conversion of vertical incidence to oblique incidence ionograms are shown in Figure 2.4, the vertical ionograms are labelled (a) with the oblique ionograms labelled (b). In the upper pair of axes, the A,B,C etc. points on the vertical ionogram translate via a transmission chart to the points A',B',C' etc. on the oblique ionogram for a 2-5

29 single propagation mode. In the second example, the critical frequency mapping is highlighted for three propagation modes. h ' I h' A A' (Q ) VERTICAL ( b ) OBLIQUE Corresponding vertical and oblique ionograms. F2 /.E /.F I / F2 (b) * / Figure Illustrations o f the conversion of vertical incidence ionograms to equivalent oblique incidence ionogram (After Davies 1965, 1969). Ionosonde waveforms There are two widely used methods for transmitting ionosonde signals, the pulsed ionosonde and the frequency modulated continuous wave (FMCW or chirp) ionosonde. Barry [1971] described the chirp ionosonde that has become commonly used for oblique incidence soundings. Here a narrow band carrier is swept across the desired range (2-30 MHz or 2-16 MHz). The receiver is time synchronised to the transmitter and maintains a synchronised frequency source that sweeps the same range as the transmitter. There is a small frequency offset between the received signal and the synchronised source, the size of this offset is proportional to the time that the signal takes to propagate from the transmitter to the receiver. This offset is plotted on the ionosonde s display for each frequency step in the swept range. The result is a graph that has an x-axis of frequency and 2-6

30 a y-axis representing the time delay between the signal being transmitted and received. Figure 2.5 illustrates the principal of operation for the chirp ionosonde, the transmitted signal is shown in blue and its frequency varies linearly with time, the received signal is shown in red and can be seen to follow a non-linear relationship with time. The distance At is equal to the propagation delay between the transmitter and the receiver, plus any system timing offset. Af is proportional to the propagation delay. The delay, At is calculated by measuring Af, this can be determined easily by measuring the frequency difference between the synthesised carrier in the receiver and the received signal frequency. Transmitted signal Received signal Time Figure The basic principle of operation for a chirp sounder. The other common waveform used by ionosondes, and the one that is typically used for vertical incidence soundings, is a pulsed signal. Here a pulse of RF energy is transmitted on a particular spot frequency and its echo recorded a short time later, this measures the time delay between transmission and reception and shows where multiple reflections occur. The spot frequency is varied with time to step across a predetermined range (i.e MHz if propagation for the full HF band is to be monitored). With such a system it is possible to monitor a few specific spot frequencies rather than sweeping a large section of the band as with a chirp sounder. 2-7

31 Dedicated ionospheric probes The techniques discussed in this section employ special signals that do not carry any useful communications data, and as such, may prevent data communication taking place whilst sounding is in progress. These probes operate within standard bandwidth channels and replace the data modulation for the duration of their use. An alternative method for channel evaluation may be to transmit a special probe signal in addition to the normal data modulation signal. Such a probe would probably have to be transmitted at a very low level to prevent degradation of the data bearing modulation, and would also need to be readily detectable at the receiver. Several long binary codes have been identified as being potentially suitable candidates for experimental work. Coded pulses In Barker [1953] a set of synchronising patterns were described, these had the properties that (a) their autocorrelation functions resulted in a single peak with the amplitude equal to the length of the pattern, and (b) at all shifts either side of the peak, or perfect match, the correlation value was limited to -1, 0 or 1. Barker s work describes these patterns in the context of time synchronising digital data streams, these patterns have since been used in various applications including radar and in the DAMSON system which will be described later. Patterns such as the Barker sequences can be used to measure the impulse response of a system using a low peak power by spreading the input energy in time, this technique is known as pulse compression. Another class of pulse compression sequences has been described by Gottesman etal [1992] and are known as modified Legendre sequences. These sequences are not limited in maximum length like the patterns Barker described, yet when used correctly they can have similar correlation features. When a modified Legendre sequence is cross correlated with another sequence which is an exact copy of itself in all but the sign of the first symbol, the peak value is equal to the length minus two, additionally, all points either side of the peak have the value of minus one. The work by Gottesman et al was aimed at low probability of intercept radar, yet can find a use in HF channel evaluation. Many techniques have been used to try to find new sequences that can be used for pulse compression techniques. In [Hu et al, 1997], a novel approach to finding such sequences is presented. Hu used a neural network search to look for sequences that exhibit good 2-8

32 autocorrelation functions. Others have used exhaustive searches to find sequence lengths up to around 48 symbols but the work by Hu presents sequences up to 100 symbols in length. The Doppler and multi-path sounding network (DAMSON) [Davies and Cannon, 1993] is an oblique channel sounding system that has been developed by the UK Defence Evaluation and Research Agency (DERA) and others to measure a number of channel parameters. Standard COTS (commercial off the shelf) equipment was used to construct the DAMSON systems which operate on pre-selected 3 khz HF channels. Several waveforms are employed for channel sounding with timing being synchronised to GPS (the global positioning system), these give the system the ability to measure absolute time of flight, multi-path spread, Doppler spread, Doppler shift and signal strength. Two of the DAMSON sounding modes employ pulse compression waveforms, these increase the amount of energy at the receiver by increasing the transmit time whilst maintaining the time resolution that could be obtained from a single pulse. The first of these is known as the delay-doppler waveform, this uses a Barker-13 pulse as described above, and is designed to measure the channel scattering function. The other is a time of flight mode and is used to measure large multipath spreads and the absolute propagation time. A CW (continuous wave) mode and a passive monitoring mode are used for signal and noise measurements. Angling etal [1998] present measurements taken using DAMSON over high latitude paths. Within this paper the authors report measuring Doppler spreads in the range 2-55 Hz and multipath delay spreads in the range 1-11 ms. These results are related the performance of a MIL-STD A modem using a modem characterisation technique developed by Arthur and Maundrell [1997]. Angling et al conclude that when there is propagation mode support, modem availabilities range from 64% to 100% for a CNR (carrier to noise ratio) of 0 db as measured by the DAMSON system. The modem data rate employed for this work was 75 bps which is very robust, however over the more benign mid-latitude paths higher data rates are being sought after. Limitations of coded pulses The usefulness of a coded pulse for channel sounding is limited by the length of the pulse, the chip rate and the pulse repetition rate. The length of the pulse, and hence the pulse repetition interval, determines the delay spread range that can be observed by using such a 2-9

33 pulse. The pulse chip rate limits the minimum mode separation that is detectable and the pulse repetition rate limits the Doppler frequency range over which signals can be measured. The DAMSON system was designed specifically to measure the channel scattering function over disturbed, high latitude paths. The Barker-13 based coded pulse waveform used in DAMSON to measure this scattering function can be shown to be appropriate for its intended application. The pulse used is 13 symbols in length followed by a silence period equal in length to 17 symbols, giving a total coded pulse length of 30 symbols. This is modulated at 2400 symbols per second and results in a pulse repetition interval of 12.5 ms, hence covering a large time dispersion range. With a repetition interval of 12.5 ms, the pulse repetition rate is equal to 80 pulses per second, this allows a +/- 40 Hz Doppler range to be measured. Other considerations when choosing a coded pulse include taking into account the processing gain of the code used, this determines how detectable the signal is in the presence of noise. Typically, a longer code sequence is required to achieve a higher processing gain, however care must be taken to match the sequence length to the expected Doppler conditions, as large Doppler shifts will cause long sequences to become less detectable due to the phase of the received symbols shifting by more than 180 over the length of the sequence. Derived channel sounding Deriving channel parameters from existing data communications signals such as modem modulations has the benefit of not interrupting data communications. This type of technique can allow channel sounding without the need for additional equipment. Additionally, modem, high speed modem waveforms may include probing sequences for the purposes of modem equaliser training. Betts et al [1970] describe a method of pilot tone channel estimation, here low level CW tones are transmitted along with data communications in assigned channels to form a channel evaluation scheme. At the receiver, measurements are made on the tones to estimate the bit error rate. The same data signal can be transmitted on all of the assigned channels at all times allowing the receiver to decide which is the best channel to listen on. According to Betts and Darnell [1975], the parameters calculated from the reception of 2-10

34 these tones include amplitude, signal to noise ratio, phase, Doppler frequency shift and spread and multi-path spread. The waveforms described in STANAG-4285 include known reference or probing symbols in each frame of data transmitted. These are intended for modem equaliser training to gain a measure of the channel s impulse response and frequency offset. Merel [1996] suggested using the STANAG-4285 reference symbols to gather estimates of the channel impulse response (CIR) and SNR for the purposes of automatic channel selection in a future automatic radio control system. Additionally, Brown and Warrington [1999] demonstrated the use of these reference symbols in producing a delay-doppler scattergram. This technique graphically illustrates the channel s effect in both the time and frequency domains. Using channel soundings for link establishment and maintenance An early example of a channel selection system is the channel evaluation and calling (CHEC) system as described in Stevens [1968]. This was developed in Canada to enhance communications between mobiles and their associated base stations. In this system the base station transmits a short signal on each of the assigned channels in turn, the transmitted signal contains information about the interference level at the base station. The mobile unit estimates from the received signal level and the decoded interference level the predicted SNR for its own transmissions back to the base station. The channel with the highest signal to noise ratio is selected for mobile to base communications. An automatic link establishment system (ALE) as defined in MIL-STD A [US DoD, 1988] allows a group of stations to individually rank all channels that are allocated for use in terms of SNR and BER. ALE employs a method of channel sounding whereby, periodically, an 8-ary MFSK waveform and a high degree of redundancy are used to transmit a station s callsign on each of the channels in turn. In between sending soundings, stations scan all channels at a rate of 2 or 5 channels per second and listen for soundings from others. When a sounding is detected, it is decoded to yield the remote station s callsign, a measure of the channel SNR and a measure of the BER based on data included in the sounding. ALE does not attempt to determine delay and Doppler parameters. ALE is widely used in military HF communications as it simplifies the link set-up process. 2-11

35 Street and Darnell [1997] discuss the use of real time channel evaluation within a communications system that can incorporate automatic link maintenance (ALM). They highlight STANAG-5066 as being a suitable communications profile and suggest that short term variations in channel conditions are best accommodated through the use of automatic repeat request operation as provided by STANAG-5066 data link layer. However, they suggest the use of embedded channel evaluation to determine the channel s ability to support modem settings from a small library of available settings (data rate, error control etc.) to cope with longer term variations in channel conditions. It is possible to make better decisions on which channel to select or which modulation scheme to use if channel parameter information such as that which is measurable by some of the techniques introduced in this chapter is available. If modem modem hardware is used as part of an adaptive HF system [ITU 1997], it is possible to automate the changing of modulation schemes and their individual parameters such as data rate and forward error correction settings based on measured channel parameters. Should the measured parameters indicate that a channel can support a high speed modulation scheme, it would be sensible to switch to such a modulation in preference to a slower scheme if the user s communication application can make beneficial use of the higher data rate. Alternatively, should a channel begin to fail and a different, more robust, modulation scheme was available it might be beneficial to switch to that automatically rather than losing all communications. Concluding remarks For modem HF skywave digital communications, advanced channel evaluation and link maintenance techniques are required to use the available frequency allocations efficiently. It will often be necessary to have a measure of the current channel characteristics to enable the best modulation scheme to be chosen. These channel characteristics, or parameters, will include a measure of the multipath and Doppler spreads, the signal to noise ratio, and possibly information about interfering signals. This kind of information can be gathered using channel evaluation techniques. Traditionally, a very simple method of channel evaluation has been used and is known as passive evaluation or frequency monitoring. This is often a manual system where an operator scans the HF bands with a receiver identifying transmissions from the area that he wishes to establish communications with. This approach to channel evaluation relies on the 2-12

36 assumption of path reciprocity, i.e. the operator can hear a transmission on a frequency from a certain location, therefore if the operator were to transmit on that frequency a receiver near to the distant location should be able to copy it. However, path reciprocity is far from guaranteed, this may be due to interference from other signals and possible differences between transmitting and receiving systems. This method relies on an ability to determine the best available channel, i.e. that with least fading, the highest SNR and the lowest amount of interference. Dedicated, active ionospheric sounding provides the greatest amount of information about the state of the propagation medium but consumes valuable transmission time and may therefore reduce the overall throughput of a particular channel. However, the parameters that sounding provides can enhance communications by allowing better utilisation of HF assets to be made. For example, higher data rates can be employed when the channel is known to be able to support them. Wide band channel evaluation systems such as the ionosonde can return information about all frequencies within the HF band between a transmitter and a receiver, however, such techniques result in pollution of the HF spectrum and may be difficult to license. In-channel evaluation is where a probe signal is transmitted within the channels allocated to the user. Received signal strength, signal to noise ratio, time dispersion, frequency dispersion, background noise and bit error rate are parameters that can be measured. If in-channel evaluation is incorporated with the modulation scheme then there are several benefits. Firstly, the probe produces an accurate representation of channel conditions for the given modulation, secondly, transmission time need not be wasted setting up an alternative probe and thirdly, equipment costs can be kept to a minimum as the necessity for extra sounding devices is removed. Channel evaluation is intended as an aid to improve communications, not to degrade them. This type of in-channel evaluation is known as embedded channel evaluation and may be used for link maintenance (e.g. adjusting modulation parameters). The STANAG-4285 waveform contains a synchronisation symbol sequence that is normally used by the modem internally, however this sequence lends itself readily to channel evaluation purposes. Also, similarly, the MIL-STD A 39 tone waveform contains a special extra tone that can be used to monitor phase distortions as part of a channel evaluation method. 2-13

37 Unfortunately, not all waveforms provide facilities for channel evaluation, and those that do may be less than optimal for the purpose. Therefore, other techniques for gathering channel parameters whilst not breaking data communications need to be researched. Ideally, the user should not need to be concerned with the communications problems that the ionosphere and interferers present, a good link establishment and maintenance system should be able to select and track the best channel available by using appropriate channel evaluation techniques. In addition to channel selection, modulation type and parameter selection may also be automated to maintain the best communications system for a given purpose. Most existing methods for channel evaluation do not provide enough information to be of any great use with modem, high speed HF systems, as examples, the CHEC system, like the pilot tone method does not measure Doppler and delay spread and the ionosonde has no method of measuring Doppler effects. Therefore, an experimental test-bed has been designed and constructed to enable an investigation into channel evaluation techniques to take place using special probe waveforms amongst other methods. Details of the systems are given in Chapter 3 with results from comparisons of the performance of differing waveforms along with channel probes being presented in Chapter 4. Error rate derived link management techniques are investigated in detail and commented on in Chapter 5. Methods for using relevant in-channel sounding have been developed and compared with other, existing methods, this work is detailed in Chapter 6. A new and novel method of providing and using channel evaluation is described in Chapter

38 Chapter 3 - Experimental configuration Introduction In the Chapter 2 it was concluded that existing methods for channel evaluation were less than optimal for use in systems that wish to employ modem, and potentially high speed, modulation schemes. In order to undertake experimental studies, it was necessary to design and build new experimental systems. To experimentally assess the performance of channel evaluation techniques, a transmitter, receiver and a reasonable length HF path were needed. Full remote control of both the transmitter and the receiver systems was required for the duration of the experimental period. The transmitter and receiver systems needed to be flexible in terms of transmit / receive frequency and modulation. Full details of the equipment and the software systems that were designed and deployed are given in this chapter. Most of the equipment deployed for these studies was commercial off the shelf technology, but was constructed into separate transmitter and receiver systems specifically for this investigation. Common system structure The Linux operating system was used on both the transmitter and receiver systems to ensure the greatest reliability possible and to simplify the development of the software components. As Linux is a pre-emptive multitasking environment with protected memory space, the various components required to run the transmitter and receiver systems could be developed and tested readily, in a modular fashion. The use of an operating system with mature networking components as standard made the remote control requirements trivial to achieve whilst being easy and reliable to use. In addition to the programs that were executed to run the many experiments carried out during this work, device drivers were written for some of the hardware components to enable the fast low-level access needed to perform time critical functions. All the control and logging software and the necessary hardware device drivers used in this research were custom written by the author with some parts of the digital to analogue and analogue to digital device drivers being based on manufacturer supplied example code. 3-1

39 Location The transmitter was initially located at the DERA Portsdown West site near to Portsmouth, UK (1.13 west, north). After problems with the antenna arrangement there and due to restrictions on the transmitter power, the system was moved to the DERA Angle site in Wales (5.11 west, north) where it was located for the remaining duration of the investigation. During the time it was located at Portsdown West, the transmitter was only used for initial system testing as a prototype. The receiver was located at the University of Leicester (1.12 west, north). This location allowed easy storage and processing of the data files that were collected due to the proximity of the University s computing facilities. A DERA testbed system was used between the Angle and Portsdown West sites to gather some statistical data as part of the experimental work. A map indicating the locations of the two DERA sites and the University of Leicester campus is presented in Figure 3.1. The great circle path lengths from the transmitter sites at DERA Portsdown West and DERA Angle to the receiver at Leicester are both around 291 km. University o f Leicester DERA Angle, Pembrokeshire DERA Portsdown West, Fareham Figure Map o f UK indicating sites used for experimental work. 3-2

40 The transmitter The transmitter system was designed to allow a variety of waveforms to be generated and transmitted in an automated manner. Remote control was considered a necessity, as was the ability to follow automatically a predetermined schedule of various transmission types on multiple frequencies. As with any remotely deployed system, reliability was of great importance. Hardware A schematic overview of the transmitter system is shown in Figure 3.2, from this it can be seen that the major components were a personal computer (PC), a telephone modem, a GPS receiver, a digital to analogue converter, an input / output (I/O) and timer card, a sweep generator, a frequency synthesiser, an RF drive unit and an RF power amplifier. The most important links between the various components are also indicated on the schematic. Telephone modem Digital to analogue converter HF drive unit Power amplifier Personal computer GPS receiver I/O / timer card Sweep generator Frequency synthesiser Figure A schematic o f the transmitter hardware. The PC used was a high quality but otherwise standard desktop machine and was the central component of the system. This had RS232 serial connections to the telephone modem, the GPS receiver and the RF drive unit. Located within the PC, the I/O / timer card and the digital to analogue converter were connected as ISA bus peripherals. To fulfil the requirement of having remote control, a telephone modem allowed password protected access from any remote location, this modem was a standard V.34 device supporting data rates up to 33,600 bps. A dedicated telephone line was made available at each of the transmit sites for remote access purposes. 3-3

41 The global positioning system was employed to ensure accurate system timing, GPS not only provides very accurate position information, but also allows the current universal time (UT) to be read to greater than 10 ps accuracy. This particular GPS receiver provides a continuous serial data stream of time and position information which was monitored by the PC, additionally it marks the precise moment of validity of the time information by a 1 pulse per second hardware output. The programmable timer and I/O card was required in order to provide an interface between the PC and the sweep generator. The reason for this was to provide an accurate, programmable delay between the GPS pulse (which is not aligned with the second boundary) and the input trigger of the sweep generator for the ionosonde component. The serial link from the PC to the drive unit was used to control the transmit carrier frequency and RF modulation type. Audio input to the drive unit was synthesised using the digital to analogue converter card under control of the PC. The digital to analogue converter chosen allowed generation of experimental waveforms at up to 200,000 samples a second, this was an Amplicon model PC-234, the schematic illustrates that this fed baseband signals to the RF drive unit (a Racal model MA3751 which covers the entire 2-30 MHz HF band). The RF drive unit in turn fed a power amplifier (with a typical average output of around 200 watts) in parallel with the low level RF from the sweep generator that formed the ionosonde. The antenna type used was, a Barker & Williamson BWD which was a 2-30 MHz wide band folded dipole. Being wide band this removed the need for an antenna tuning unit at the cost of a small reduction in transmit efficiency. For the ionosonde transmissions, the use of a wide band antenna was essential as the frequency of operation is continually changing. The frequency synthesiser (a PTS model 310) and sweep generator formed the low-level RF drive source for the ionosonde transmitter function. Here the sweep generator was a device that when triggered would program the frequency synthesiser with the correct frequency for the current position in the ionosonde cycle. The FMCW ionosonde transmit format for this system dictated that the frequency would be swept from 2-30 MHz in 1 Hz steps at a rate of 100 khz per second. 3-4

42 Software As mentioned above, dedicated software was developed to control the transmitter system. Each program and driver was written as a sub-project of the entire system to allow them to be developed and tested as independently of each as possible whilst retaining shared functions and data centrally. Figure 3.3 is a schematic of the programs and device drivers that were written to make up the transmitter system. At the heart of this diagram is the Linux operating system which is responsible for allowing the user level programs (i.e. regular applications) to be multi-tasked as necessary, these are shown on the left hand side, and for providing the stricter kernel level environment for the device drivers, on the right hand side, to operate in. The user level programs access the hardware devices via kernel calls to the device drivers, this is achieved using a common interface that UNIX-like operating systems present. The device drivers work as add-on parts to the generic operating system kernel and provided a structured approach to accessing and controlling specialised hardware. Details of the purpose and functionality of the transmitter s various software components are given below. User space Kernel space GPS daemon Ionosonde transmitter daemon Waveform transmitter scheduler Linux operating system Input / output card device driver D/A converter device driver Figure 3.3- Schematic o f the transmitter software. The I/O and timer card driver allowed this card s memory registers to be programmed in order to introduce a logic delay and a binary switch as described previously in the hardware section. This driver was called by the ionosonde transmitter daemon when it required to program the necessary timer delay between the GPS pulse and the start of the 3-5

43 ionosonde sweep and when the output of the timer was to be connected to the sweep generator trigger input. To control the digital to analogue converter card, the digital to analogue converter driver programmed various registers on the card to set the appropriate mode of operation, the card itself was very flexible and could be used in many applications, in this case it was required that it operated at a fixed sample rate using an external clock taking blocks of samples from a memory buffer and that it used hardware interrupts to signal when more samples were required. The device driver maintained a kernel-level buffer of samples that were taken as required from the user-level waveform transmitter scheduler application for as long as they were made available. The GPS serial data stream was continuously monitored by the application called the GPS daemon, this received the messages passed from the GPS receiver and translated them to time and position information. Rather than just setting the PC or operating system clock based on this information, which in tests proved to result in poor accuracy, the GPS daemon monitored a number of software FIFOs (first in, first out buffers used to provide a form of inter-process communication) and fed the time and position information to any that were linked to other applications. This allowed programs such as the ionosonde daemon and the waveform transmitter scheduler to receive this information with good accuracy. In order for the ionosonde function to work correctly, the chirp sweep must be started at fixed intervals with an accuracy of better than one millisecond. This system was designed to begin the ionosonde sweep on a second boundary. To achieve this, the ionosonde daemon monitored the GPS time information via a FIFO and at a short period before the sweep was due to start, the programmable timer was set to provide a delay between when the GPS pulse would arrive and the next second boundary, this was achieved with a retriggerable monostable timer. The output of the monostable provided the input to a switched line on the output section of the I/O / timer card, this was only enabled after the last GPS time message prior to the sweep start was received. Although the offset of the GPS messages from the true second boundary varied, the rate of this was less than one millisecond per minute and hence the timer could be programmed quite well in advance of the sweep start. The waveform transmitter scheduler program was written to automate the sending of sounding waveforms. This program read the experimental schedule from a text 3-6

44 configuration file which specified waveform filenames, waveform sampling rates, plus times and frequencies for transmission on. The waveforms were stored in binary files on the PC, having been created earlier either on the PC or remotely. The format for the data was to use 16 bit unsigned integer samples at any sampling rate up to 200,000 samples per second. The waveforms specified in the configuration file were loaded into memory when the program was started, the program then monitored the time information supplied by the GPS daemon waiting for the next event in the schedule to become due. The events that could be specified in the schedule include setting the transmit frequency and sending a particular waveform. The schedule could be repeated either a fixed number of times or continuously until manually stopped with the repeat cycle being every day, hour or minute. Other functions such as handling incoming modem connections and general system housekeeping were standard parts of the operating system and were used as supplied. Remote control was achieved by setting up a remote network connection using the point to point protocol (PPP) over the telephone modem, over this, telnet and ftp were used for remote terminal emulation and file transfers. The receiver The receiver system was designed as a complement to the transmitter system described above, and was constructed in a similar manner. The main design criteria were that the system could tune to any frequency within the HF band, have sufficient bandwidth for the modem waveforms of interest, be able to run automatically following a time and frequency schedule. Also it was required that the system would record oblique incidence ionograms as received by an ionosonde receiver. Hardware Figure 3.4 is a schematic diagram of the main components of the receiver system, these were a PC, a GPS receiver, an HF receiver, an analogue to digital converter, an ionosonde receiver and a data storage device. 3-7

45 Analogue to digital converter HF receiver Personal computer Data storage Ionosonde receiver Offline processing GPS receiver Figure The receiver system. The PC was a high specification tower system and like the PC of the transmitter system, it was the core component of the entire receiver system. RS232 serial connections linked the PC to the HF receiver, the GPS receiver and the ionosonde receiver. The analogue to digital converter was present as an ISA peripheral card inside the PC. The data storage element was realised by using an ethemet network connection to a remote hard disk and DAT (digital audio tape) drive1. The GPS receiver was used as described for the transmitter to provide accurate timing information. The HF receiver used was a Racal model RA-6790/GM, this covered the 2-30 MHz HF range in USB, LSB, AM, FM and CW modes and was fitted with 3 khz IF filters for this particular experimental work. The receiver was remotely controlled by the serial link from the PC in order to set the operational frequency and mode of demodulation. Baseband audio output from the receiver was sampled by the analogue to digital converter card in the PC. The analogue to digital converter was an Amplicon Liveline model PC-226, this was capable of taking up to 50, bit samples per second in single channel mode. A sampling rate of 9600 samples per second was chosen for most of the experimental work 1The DAT system can be used to record, digitally, audio signals, but in this system it was used to store computer data files in a portable and accessible format. DAT is commonly used to archive computer information owing to the high capacity and quoted reliability of the tapes used. 3-8

46 as this was in excess of the Nyquist rate (to ensure faithful reproduction of the baseband signal) and provided an integer number of samples per symbol for transmit symbol rates of 2400 baud. To collect oblique ionograms the equipment used was a BR Communications RCS5A chirp-sounder receiver. This device displayed ionogram plots as they were received on an internal CRT display, additionally it presented the ionogram data as an RS232 serial data stream to allow computer storage of the plots. Software The main elements of the software written for the receiver system are shown schematically in Figure 3.5. The relationship between the application programs and the device drivers was the same as for the transmitter with the programs necessary to execute the experiments being run in the user-space environment and the device drivers in the kemel-space. Details of the individual software components are given below. User space Kernel space GPS daemon Ionosonde receiver daemon Linux operating system A/D converter device driver Audio receiver scheduler Figure 3.5-A schematic o f the receiver software. The device driver for the analogue to digital converter provided the interface between the receiver scheduler program and the hardware registers and memory on the card. The card was used in a mode that gathers samples continuously for as long as the driver requests them. The driver buffered up to 16,384 samples and notified the calling program when 3-9

47 there were 1024 or more samples ready for collection, this buffering prevented samples being lost whilst writing to the hard disk. The GPS daemon was the same as described for the transmitter. The ionosonde receiver daemon monitored the serial port to which the RCS5A chirpsounder receiver was connected and stored the ionogram plots as they were received. Each plot was saved to a separate file named according to the current UT date and the ionogram sequence number. Utility programs were also written to redisplay the ionograms on computer monitors and to allow them to be used in documents. The audio receiver scheduler program was written to automate the receiving and sampling of off-air signals (i.e. the sounding waveforms). Like the waveform transmitter scheduler, this used a text configuration file to allow the user to set the sampling rate, receive filename, time of event and radio frequency to tune to. The schedule also allowed specific archive commands to be executed at given times to store the large quantity of data such a system was likely to collect, these commands were used to move the data files from the local hard disk to a remote network drive ready for later analysis or tape storage. The audio sampling function was designed to run continuously, for as long as scheduled, this was to enable very large data sets to be collected if required. Validation o f equipment The equipment described above was thoroughly tested to validate the reliability and functionality. During the testing period a transmitter to receiver frequency offset was noted and further investigated. This offset was found to be fixed for all assigned frequencies and was also stable with time and temperature, however it was decided to perform calibration tests between our transmitter and receiver systems periodically during the data collection campaign. The channel evaluation waveforms that were to be transmitted may contain no clear centre frequency peak at the receiver due to their modulation types, so a single tone was transmitted as the calibration signal. At the receiver s baseband output, the frequency spectrum clearly showed the tone and any adjustment values could be calculated easily. To prove the systems, schedules for the transmitter and receiver were written to transmit and record a sounding waveform. The sequence was to transmit 30 seconds of Barker-13 probe signal every five minutes changing between five separate frequencies. At the receiver 30 seconds of baseband audio was sampled for each frequency and stored for 3-10

48 offline processing. Two examples of Barker-13 sounding plots are presented in Figures 3.6 and 3.7. The first example shows that the Barker-13 sequence was detected in the presence of background noise, the second shows the sequence detected with very little background noise. These initial tests proved successful and the probe results indicated varying channel conditions across the day as might be expected. A more intensive and useful period of system evaluation was undertaken and is described in the following chapter. Barker-13 sounding: MHz 12/11/ :30 UT Delay (m s) Amplitude Figure Barker-13 sounding scattergram for Angle - Leicester path at 15:30 on 12/11/1996 at 4.92 MHz. 3-11

49 Barker-13 sounding: MHz 12/11/ :30 UT Delay (m s) 40 TT 20 x >-> g 0 a> Amplitude Figure Barker-13 sounding scattergram for Angle - Leicester path at 19:30 on 12/11/1996 at 4.92 MHz. 3-12

50 Chapter 4 - Comparisons of different modulations over the course of several days In Chapter 1 several different modulation schemes were introduced which collectively support a large range of data rates. These differing modulations and data rates have widely varying performance characteristics when used over the variable channels that HF provides. Owing to this, an appropriate modulation should be chosen to suit the propagation path conditions and the nature of the data to be transmitted. As an example, naval shore to ship broadcasts are transmitted all over the world via HF and are genuine one-to-many (one-way) communications. These broadcasts need to be received reliably since there is no protocol in place to request a retransmission. Conversely, the quantity of data to be sent is quite small and does not require a high data rate for reasons of timeliness. A 75 baud FSK modulation is typically used for these communications and satisfies the operational need. Another example is the necessary use of higher speed modulations for interconnecting computer networks via HF over reasonably benign paths. Here a minimum speed of 1200 bps is usually required for the communications protocols to function, but faster links are very desirable. A 1200 bps 8-ary PSK waveform such as STANAG-4285 is appropriate as the slowest choice, however, if the channel will support a faster waveform, then one should be used. To investigate the effects which the HF channel has on various different modulations, a number of on-air experiments were performed. These experiments were designed to compare the performance, in terms of the error rate and the data throughput, of differing modulation schemes working under the same propagation and interference conditions. Measures of the propagation conditions were also recorded using two of the methods introduced in Chapter 2. Experimental configuration Without resorting to simulated channels or modelled performance it is not possible to find two allocated, non interfering channels over which to run simultaneous transmissions using the same antenna equipment, therefore for this work a close compromise was chosen. The transmission format chosen, as depicted by Figure 4.1, was one where the modem waveforms under test were sent in sequential bursts each lasting for 150 seconds, these 4-1

51 data bursts were preceded by a Barker-13 probe sent for sounding purposes which lasted for 15 seconds. A further period of 15 seconds was sampled after the modem waveforms had finished out of interest in the background channel noise and interference. This could be replayed and listened to manually if required. Probe waveform (15s) Gap (15s) Modem waveform (150s) Figure Form at o f waveform transmission burst. Using the equipment described in Chapter 3, four different modem waveforms were sent in bursts during each transmission period, an illustrative example of this is shown in Figure 4.2. There were two transmission periods per hour, as depicted in Figure 4.3, in order to capture a reasonable set of conditions across the day. STANAG bps MIL-STD A Serial tone 1200 bps MIL-STD A 39 tone 1200 bps BFSK 600 bps Figure Example form at o f sequential waveform bursts fo r one transmission period. Transmission period 1 (start time 1 minute past the hour) Transmission period 2 (start t ime 31 minutes past the hour) No transmission (18 minutes) No transmission (18 minutes) Figure Form at o f transmission periods that were repeated each hour. The waveform bursts sent were the audio output of a Rockwell MDM-3001 HF modem, these were generated from known data sequences which each consisted of 400 bytes of pseudo-random data. At the receiver, the sampled audio signal was demodulated by 4-2

52 replaying it through a similar modem. The output stream of bytes was stored to a file and later examined to determine how many bytes were received correctly. This was achieved by dividing the known sequences into small segments and searching for each segment within the stored data. The Barker-13 probe waveform that was sent during each transmission period was processed in order to form a measure of the channel scattering function, more detail is presented on this technique in Chapter 6. The most important parameters that can be measured using this technique are those of delay and Doppler spreads as these have a clear influence on the performance of a modulation scheme. The delay and Doppler spreads encountered during the experimental period were extracted from the Barker-13 soundings using a technique similar to that described in Angling [2001]. This automated technique aims to identify the time boundaries of detected propagation modes before producing a Doppler profile to allow the frequency domain boundaries to be identified. After gaining these time and frequency domain boundaries, statistical analysis is used to produce an estimate of the respective spreads. To allow a quick estimate of the channel SNR to be made, an automated technique was developed which also made use of the results from the Barker-13 soundings. This estimate was based on the proportion of power estimated in the received coded pulse compared with the background noise and any interferers present, this measure of SNR was called the ESNIR (estimated signal to noise and interference ratio). The output from the Barker-13 probe technique was introduced in Chapter 3 and an example was illustrated in Figure 3.6. It is from the top pane of this type of plots that the ESNIR values were calculated. The ESNIR number is a ratio of the power contained by the probe response and the background noise level, see Figure 4.4 for an idealised example of single moded propagation. The power from the probe sequence is considered to be contained within the triangle of height h and base b, h is normalised to 1 by the processing method. For the simple estimates of SNR that this technique aimed to give, a received signal power was assumed. Based on the number of propagation modes commonly observed over the Angle to Leicester path (typically single or dual moded), 1.5 was chosen as the average number of modes and used to form the estimated received signal power. The dimension b for a non-spread signal was typically around 1 ms, hence the chosen value for received signal power was 0.75 taking in to account the average number of modes. The average noise level 4-3

53 is indicated by the dimension n. To form an estimate of the noise power, this level was multiplied by the transmit time of the probe signal, 5.4 ms. Therefore, the ESNIR is given by Equation 4.1. ESNIR = 10 * log10 5.4*«(4-D The ESNIR calculation can be seen to have a floor value of about -8.5 db due to the estimate of the received signal power and the maximum noise power measurable. This has the effect of suggesting there is always some signal present even if the noise level is measured at its maximum value. For the purposes of the results presented here, these estimations and this floor level are quite adequate. i i Sounding sequence peak Background noise Figure Example o f values used to form the ESNIR estimate fro m Barker-13 sounding results. Comparisons of modulation types Data sets collected over six full 24 hour periods during April and June 1997 using a frequency around 5 MHz over the Angle - Leicester path are presented below. Each 24 hour set is a comparison of two different waveforms. In each case the waveforms were of different data rates and in some of the cases, different basic modulation types were used. The sunspot number (SSN) observed during this period of data collection was around 13 indicating a fairly low level of activity. Using this SSN as an input to the VOACAP prediction program, it was estimated that propagation at around 5 MHz should take place between around 07:00 and 22:00 hours UT. The geomagnetic indices for the six days presented had ap values of typically 2-6, the only exceptions to these are for the darkness 4-4

54 hours on 28/6/1997 and 29/6/1997 where ap extends between 9 and 15. Therefore the data presented here was collected under non-disturbed conditions. Data collected on 26/4/1997 The modulation types from this day that are compared here were both STANAG-4285 with one operating at 1200 bps and the other at 2400 bps. A summary of the day s results is presented in Figure 4.5, which is made up of five separate plots described individually below. The first two plots in this figure are of the measured data throughput for the two modulation types. One step is plotted for each half hour period of the day relating to a single transmit period. The values of the steps and the calculated mean figures indicate that the 2400 bps waveform outperformed the slower waveform. Over the course of the day, the 2400 bps waveform had an average throughput that was 41% greater than the 1200 bps waveform. If both modems had been performing without error it would be expected that the throughput for the faster modem would have been 100% greater throughput than that of the slower modem. The third plot is a graphical comparison of the relative throughput performance of the two waveforms. The green line, labelled EquaT\ indicates equal data throughput performance. The magenta line, labelled Ratio, indicates the relative performance expected of the modems if both were operating at maximum throughput (i.e. the faster modem would yield a throughput twice as great as the slower modem). The scale of the y-axis above the Equar line represents how many times greater throughput was achieved by the faster modem over the slower modem. Below this line, the axis represents how many times greater throughput is achieved by the slower modem over the faster modem. Given a choice, the modem to use would be chosen based on which side of the Equar line the curve lies. If the curve lies on the upper side of this line, then the faster modem should be chosen as it is outperforming the slower modem, if it lies on the lower side of this line then it is the slower modem that has a greater throughput and hence should be chosen. For the majority of this day s comparison the curve lies close to, but above, the Equar line. This indicates that the two modems were operating comparably with the higher speed modem just outperforming the slower modem. The fourth plot in this figure is the estimated signal to noise and interference ratio curve, although this is not a direct measure of SNR it does indicate how the detected power in the 4-5

55 received probe signal varies across the day with respect to background noise and interferers. Where the ESNIR value is near to -8.5 db, very little or no signal is detected by the algorithm used. The mean value indicated in the figure is the mean of the individual linear signal to noise ratios that has been converted into decibels to match the graph axes. The ESNIR values for this day indicate several periods where the signal levels are good and these generally correspond to good recorded throughput values. The final plot is an estimation of the number of propagation modes supporting transmissions at the frequency of operation taken over the path used. The number of these modes was estimated from ionograms collected over the day. Based on the ionograms captured nearest to the transmission time, the number of visible modes was counted and recorded. These figures were then plotted against time of day to indicate when propagation was detected by the ionosonde and if it was, then by how many modes. The results of this plot in conjunction with the throughput plots for this data set indicate that successful data transmission took place during most of the times that the ionograms indicated propagation was possible (as expected). The underlying Doppler and delay spread characteristics of the channel were measured from the results of the Barker-13 probing using the automated method described earlier in this chapter. Figure 4.6 contains both the Doppler and delay results as functions of time for this day. In the upper pane of this figure, a non-spread single moded channel will result in an effective multipath spread value of zero. A spread or multi-moded channel will have an effective multipath spread value based on the relative amplitudes of the received modes and their spread widths. The Doppler spread is a measure of the frequency range which contains the majority of the probe power. From this figure, the delay spread is small at all times, with only three periods exceeding 0.5 ms. The Doppler spread measured was more variable than the delay spread, with a range between just under 1 Hz up to around 4 Hz at the extreme. When the delay and Doppler spreads could not be estimated from the given data, no points were plotted. To quantify the interference background, the frequency spectrum of the received samples was examined for the duration of each transmit period across the day. From these spectra the number of strong and weak interferers within the bandwidth required by a receiving modem was estimated and plotted. A strong interferer was defined as a signal with a bandwidth much greater than about 2 Hz and an amplitude greater than the average 4-6

56 amplitude of the modem signal. Typically, interferers classified as strong had bandwidths greater than about 20 Hz and were clearly visible in the presence of the modem signal. A weak interferer was defined as a signal that either had a very narrow bandwidth (i.e. less than 2 Hz) or one that had a wider bandwidth with a low amplitude (i.e. less than that of the modem signal). A lot of weak interferers were signals that were much less than 1 Hz wide or were only readily detectable when the modem signal was not being transmitted. The estimation of interferers for this day is plotted in Figure 4.7. In the case of the STANAG-4285 waveforms used on this day, the entire 3 khz bandwidth was considered to be required by the demodulating modem and therefore any interferers evident at all were counted. On this day, there were a number of weak interferers present at most times but these appear to have had little or no effect on the performance of either of the data waveforms. Strong interferers were present during several of the transmission periods where propagation conditions should have allowed good throughputs. On only one of these occasions did there appear to be any correlation between the presence of a strong interferer and a significant drop in data throughput. This loss of throughput affected both modems similarly. On this day, both waveforms performed reasonably well. The number of strong interferers was small and the nature of these had little affect on the data throughputs. The channel conditions were reasonably benign. The multimoded propagation that was evident for much of the day appears to have had little effect on the performance of waveforms, this is due to the delay spread not exceeding the 5 ms tolerance that these waveforms typically have. 4-7

57 STANAG bps 100 Mean STANAG bps B Mean Relative data throughput performance Estimated signal to noise and interference ratio Mean Estimated number of propagation modes Hour(UT) 20 Figure Data throughput, ESNIR and estimate o f mode count results for 26/4/

58 I Hour (UT) Figure Effective delay spread and Doppler spreadfor 26/4/1997. STANAG bps 6 Weak interferers Strong interferers Z 4 I* o a o. moo i <p * * X G G 0 - O < i } cb cb» 10 «o 15 - I 4-20 >O0OG<) STANAG bps Weak interferers Strong interferers Z Q G O G G -6 <j> G O * i i. «-» - 4 ( 5 G G -0G ^ 0< 5K i) 10 -x-* - A-i Hour (UT) 15 cpo cp G-rp cp h& - G- 20 * oo-cxp 0 - x -f Figure 4.7 Number o f evident interferers for 26/4/

59 Data collected on 27/4/1997 Two STANAG-4285 waveforms were transmitted on 27/4/1997, one at 1200 bps, the other at 2400 bps. The processed results are presented in Figure 4.8. The throughput results for this data show that both modulation schemes were able to pass data for the majority of the day with consistent throughput figures. The 1200 bps waveform performed quite well during the hours of propagation, however, the faster waveform was not able to achieve twice its throughput and hence was not operating as efficiently. On the throughput comparison plot, as in the previous example, for most of the day the curve lies just above the Equar line. This indicates that the higher speed modem marginally outperformed the slower modem in terms of throughput. The multimoded propagation that was evident for much of the day had little effect on the performance of the STANAG-4285 waveforms as the delay spreads recorded were small. The propagation mode counts at around 09:00 and 15:00 hours were estimated to be zero yet data was still passed, this was due to ionograms at these times not being received correctly. From Figure 4.9, the channel delay and Doppler estimates indicate very little effect from multipath delay but the Doppler spread was often above 3 Hz which is likely to have precluded the use of any higher speed waveforms. Both modem waveforms performed well over this day owing to fairly high ESNIR values and hence good signal levels and the almost complete lack of strong interferers (see Figure 4.10). The number of weak interferers was low throughout the majority of the time where good propagation was evident. 4-10

60 150 STANAG bps I 2 Mean STANAG bps & S CL I 50 n Mean Relative data throughput performance 20 Ratio 2 Equal 1 Estimated signal to noise and interference ratio co "O Z CO w -10 Mean Estimated number of propagation modes 20 1 l L i L Hour (UT) Figure 4.8- Data throughput, ESNIR and estimate o f mode count results for 27/4/

61 Hour (UT) 20 Figure Effective delay spread and Doppler spreadfor 27/4/ tr Q(!) STANAG bps Weak interferers Strong interferers I I OO ( 3<M< 10 3<j) 30 > I 63 I 3 3 O O O 33 **33 h STANAG bps 1 Sc i I ED z i ooao Weak interferers Strong interferers 0 3 J *» - * - «-» - * <» o «3-3- Hour (UT) Figure Number o f evident interferers for 27/4/1997.

62 Data collected on 13/6/1997 The third set of data was collected on 13/6/1997 and consisted of two STANAG-4285 waveforms operating at 1200 and 2400 bps. The main processed results are presented in Figure It is clear from the throughput curves of the first two plots in this figure that the slower rate waveform outperformed the higher rate waveform by a considerable margin. In fact, the higher rate waveform performed particularly badly on this day and only managed to achieve a projected average throughput rate of 6 bytes/second. Even though the 1200 bps waveform outperformed the 2400 bps waveform by more than a factor of three, it did not perform particularly well and the throughputs measured for the individual transmission periods varied widely. A comparison of the relative throughputs (the third plot in this figure) shows several periods where the 2400 bps waveform did actually perform better than the 1200 bps, but for the majority of the time it echoes the throughput results mentioned above. The last two plots in this figure shows that the propagation path on this day suffered from multiple propagation modes, and also from a variable ESNIR which indicates that the signal or noise / interference level varied greatly at the receiver. The variability of the ESNIR appears to have affected both waveforms and is likely to have been caused by fading. To examine this possibility further, the received baseband samples were played back through a loudspeaker, these revealed the characteristic sound of a slow fading signal. A representation in the form of a spectrogram of the channel frequency spectrum for a period of time when fading was evident is presented in Figure 4.12, darker shades in this figure represent stronger signal levels. The light shaded diagonal stripes are evidence of slow fading where the frequency bands affected by this fading vary with time, this is characteristic of interference fading caused by multipath propagation. The modem signal was sent in short bursts and hence the lighter shaded horizontal bands are times when the modem signal was not being transmitted. The Doppler and delay estimations in Figure 4.13 indicate a reasonably high level of Doppler spread across most of the day with spreads of 3 Hz or greater being common. Such Doppler spreads in conjunction with any spreading in the time domain will have caused demodulation problems for the 2400 bps waveform. 4-13

63 150 STANAG bps o IXI Mean STANAG bps r 1 ' 'T 20 I s CL f 50 E - n 1 MH., 1 1! Relative data throughput performance - k - H - Mean 21 Ratio 2 Equal Estimated signal to noise and interference ratio Mean Estimated number of propagation modes Hour (UT) Figure Data throughput, ESNIR and estimate o f mode count results for 13/6/

64 WsSs^'^v* Li/Mijfaa Frequency (Hz) Figure Spectrogram taken around 14:30 on 13/6/ Q. r. CL Hour (UT) Figure Effective delay spread and Doppler spreadfor 13/6/

65 The number and category of interferers detected on this day are presented in Figure There appears to be little correlation between detected interferers and poor throughput on this day. In general, on this day, poor SNR conditions were likely to have been responsible for the poor performance of both waveforms. In addition, the higher speed waveform is likely to have been adversely affected by its low tolerance to Doppler spreads. STANAG bps Weak interferers Strong interferers i 4 I I* - 9 f *»-*-«f J : *-* * * ooo'o<!> 0 * I * o -o oo OO i-o-e-o oo09" 10» 9 9 e (WmooG»* oo-o STANAG bps Weak interferers Strong interferers i 4 I 2 -#00-01 «09 <{> *» j : - - ooocw **4 - o-o Hour (UT) Figure Number o f evident interferers fo r 13/6/1997. Data collected on 28/6/1997 Another example of a slower data rate waveform outperforming a potentially faster waveform was discovered after analysing the data collected on 28/6/1997. The two waveforms compared in this example were 600 bps FSK and 1200 bps MIL-STD A. The throughput, ESNIR and propagation mode count results for this day are presented in Figure It is interesting to note that the FSK modem has no forward error correction or equalising capabilities, yet on this day resulted in an average throughput of twice that of the MIL- STD A serial tone waveform. This is indicated by the mean throughput values in the first two plots of this figure. 4-16

66 In contrast to the previous examples, data throughput was achieved for the entire 24 hour period on this day, and remarkably, it was the poorer overall performing waveform that achieved this. Therefore, the relative data throughput plot shows quite correctly that the slower waveform was the one of choice for the majority of the day, yet during the early hours of the morning, the faster (and probably of greater significance, more complex) waveform was shown to be the better performer. The ESNIR plot shows that the estimated signal to noise and interference ratios were low but variable across the day. Similarly, the number of modes estimated from the ionograms varied between being single moded and multimoded (two or three modes in this case). There were three periods when no propagation could be detected using the ionograms, the first two of these corresponded to the FSK waveform failing. The third period corresponds to a time when traces on the ionograms were not received. Figure 4.16 indicates little effective multipath spread was detected from the Barker-13 sounding results and that the Doppler spread was mostly below 2.5 Hz during this day. Based on these results, it is likely that neither Doppler nor delay spread were responsible for the poor performance of the 1200 bps waveform. The results from analysing the types and locations of interferers is presented in Figure 4.17 and shows that for the 1200 bps PSK waveform one or more strong interferers was present for the majority of the day. However, the FSK waveform did not suffer from any strong interference near to the two tone frequencies. The interferers that affected the PSK waveform were still present within the 3 khz channel bandwidth but were sufficiently distant from the FSK tone frequencies to cause little or no effect on the performance of this waveform. It appears that low SNR due to the strong interferers present was responsible for the poor performance of the MIL-STD A waveform. Also, the SNR requirements of the two waveforms used on this day are reasonably similar, but the surprising performance of the MIL-STD A waveform at very poor signal levels was probably due to the frame synchronisation method employed. This would have allowed the receiving modem to detect a signal in the presence of a large amount of noise and align the data bits received correctly, the FSK waveform used a less robust technique. 4-17

67 150 cl S CL sz MIL-STD A 1200 bps Mean BFSK 600 bps 150 m a 100 s B5 50 S Mean Relative data throughput performance 20 Ratio 2 Equal Estimated signal to noise and interference ratio CD ;o a: 2 W w -10 Mean Estimated number of propagation modes Hour (UT) Figure Data throughput, ESNIR and estimate o f mode count results for 28/6/

68 a. 3 a. «a Hour (UT) Figure Effective delay spread and Doppler spread fo r 28/6/1997. MIL-STD A 1200 bps Weak interferers Strong interferers t Offi <->00 Q-QQ< I«$ I f 20 5 O O O O O d )(W>O O O * BFSK 600 bps Weak interferers Strong interferers.>oo<->oooa oo-ooo (j> a- o-o a-o.*>c* fo $» <s>(i) 1«) 0 0 0t>0 (ih - Hour 0 ao Figure Number o f evident interferers for 28/6/

69 Data collected on 29/6/1997 The waveforms analysed from 29/6/1997 were 600 bps FSK and 1200 bps STANAG The two waveforms resulted in similar values for the average throughput, as indicated by the first two plots in Figure Of greater interest than either the fact that the throughputs of the two waveforms were similar or that the ESNIR for this day was low and variable, as indicated in the third plot of the same figure, is that for a fairly long period, between 08:00 and 11:00 hours the FSK waveform outperformed the PSK waveform considerably. This anomaly warranted further investigation and the frequency spectrum was analysed to try to determine the reason for this. For the transmit period around 08:00 the received samples relating to when the FSK waveform was sent were processed using an FFT algorithm to determine the frequency spectrum. These results can be found in Figure Around 08:00, FSK demodulation was successful and yielding over 40 bytes/second in throughput terms. From the frequency spectrum plot, the two FSK tones are found centred around 1575 Hz and 2425 Hz (the waveform used was a 850 Hz wide-shift FSK employing a centre frequency of 2000 Hz), however, there are also two interfering signals with much larger amplitudes indicated on this plot at around 31 and 826 Hz. Figure 4.20 is a similar frequency spectrum plot taken from the received samples corresponding to the transmission of the PSK waveform, again the two interfering signals are clear. These interfering signals have little effect on the FSK waveform as they are quite distant from it, but would have dramatic effects on a PSK waveform that requires most of the 3 khz channel bandwidth. Further analysis of the frequency spectrum for this day reveals that the interferers are present, but varying in amplitude up until around 19:30. At this time the frequency spectrum, as presented in Figure 4.21, is much cleaner showing the FSK tones at a high relative amplitude and little evidence of the interferers. Figure 4.22 is a similar frequency spectrum plot showing the PSK waveform with no evidence of strong interference. 4-20

70 150 I S. 100 I 50 STANAG bps Mean & s. 100 f 50 i BFSK 600 bps Relative data throughput performance PL.] Mean 14 Ratio 2 Equal Estimated signal to noise and interference ratio Mean Estimated number of propagation modes 01 J Hour (UT) 20 Figure Data throughput, ESNIR and estimate o f mode count results fo r 29/6/

71 FFT analysis: 29/06/ :00 UT a> % E < Frequency (Hz) Figure Frequency spectrum fo r 29/6/ :00 UT during FSK transmission. x 10 FFT analysis: 29/06/ :30 UT Frequency (Hz) 3000 Figure Frequency spectrum for 29/6/199719:30 UT during FSK transmission. 4-22

72 FFT analysis: 29/06/ :00 UT Frequency (Flz) Figure Frequency spectrum fo r 29/6/ :00 U T during P SK transmission. x 105 FFT analysis: 29/06/ :30 UT Frequency (Hz) Figure Frequency spectrum fo r 29/6/199719:30 UT during P SK transmission. The analysed Barker-13 sounding results presented in Figure 4.23 indicate low effective multipath spread and Doppler spreads mostly around 2 Hz. It is worth noting that there 4-23

73 were several periods across the day when analysis of the Barker-13 soundings failed to produce estimates for the spreads. These periods generally match times of very low or zero data throughput as indicated in Figure Poor propagation conditions are the most likely cause for these outages and the propagation mode counts taken from the ionograms reflects this. Figure 4.24 reveals a large number of weaker interferers being present on the frequency used during this day. For a lot of the day one or two strong interferers are detected. When two strong interferers are noted, a significant loss in throughput from the PSK waveform compared with the FSK waveform is resultant. During the times when a single strong interferer is present, less effect is seen on the PSK waveform s throughput in comparison with the FSK waveform. CO E 3.5 g 2.5 Q..C c5 Q. 13 E I Hour (UT) Figure Effective delay spread and D oppler spread fo r 29/6/

74 82 6 ' * 9 QQ0O SO <1> - - 0G STANAG bps -0 - j i i I Weak interferers Strong interferers 0-0 Cj> j Q~& a 6! T? *" &&«-»- 6!! I oa dr-a 6 > $ l *-A.» * 0 <fe««-* «I i * *--* k A 89 0() BFSK 600 bps 82 6 Weak interferers Strong interferers b d>c> i « o oe-o o e H 9 10 (p 9 00 e- * - 4 k Hour (UT) Figure Number o f evident interferers fo r 29/6/1997. Data collected on 30/6/1997 The data collected on 30/6/1997 consisted of a 600 bps FSK waveform and a 1200 bps STANAG-4285 waveform and is summarised in Figure 4.25 using the same format as the previous examples. From the mean throughput values of the first two curves it is apparent that the higher speed waveform achieved an average throughput of almost double that of the slower waveform. The times at which the higher speed waveform was outperforming the other are clearly shown in the relative throughput plot where the curve lies on the upper side of the Equal line. Where the ESNIR plot indicates a level greater than about -8.5 db, some passage of data was achieved with both waveforms in general. The number of modes counted from the ionograms collected over this day varied between one and two, with singled moded propagation being supported for most of the day. There were four periods where no propagation was detected by examining the ionograms and these all relate to times when no or very little data was passed across the link 4-25

75 Like the first example, using a higher data rate waveform on this day has resulted in an increased total data throughput achieved compared with the slower waveform. Also for most of the individual transmission periods, the higher rate waveform has proved to have superior performance. Neither waveform has worked to its full capacity but using the higher rate waveform would have been advantageous. The outage at around 18:30 hours, as indicated by both throughput plots, the ESNIR plot and the number of propagation modes counted from ionograms was probably due to a loss of all propagation over the path or a strong wideband interferer local to the receiver. From the Barker-13 sounding plot produced from samples taken at this time there was no evidence of a peak of any magnitude, see Figure In the poorest of SNR conditions it is usual to see some trace of the probe waveform on the results plot due to the sequence s high processing gain. When the samples taken at 18:30 were reproduced through a loudspeaker, there was no evidence of any interferer or even a signal of any type. The effective multipath spreads observed on this day as presented in Figure 4.27 are generally very low. The only significant spreading occurred around 07:30. The data throughput was negligible at that time too, but it is most likely that this was due to poor propagation conditions instead as the previous few transmission periods, with no significant spreading, also had very poor throughputs and similar ESNIR values. In general the Doppler spreads are very low on this day with the exception of a measured spread of around 3.8 Hz around 07:30. Figure 4.28 indicates that there were few strong interferers detected on this day although the count did reach three for the 1200 bps waveform during the intervals around 22:30 and 23:00 where the data throughput was low. During the same intervals, one strong interferer to the FSK waveform was detected and this waveform s throughput was also low. Again, the most likely cause for poor throughputs at this time would have been poor propagation support. 4-26

76 150 STANAG bps 100 Mean BFSK 600 bps I S- 100 S Q. f 50 S Relative data throughput performance 20 Mean 13 Ratio 2 Equal Estimated signal to noise and interference ratio co 33 a: z w w -10 Mean Estimated number of propagation modes 20 I * o Hour (Ul) 20 Figure Data throughput, ESNIR and estimate o f mode count results for 30/6/

77 Barker-13 sounding: MHz 30/06/ :30 UT 4Q Received power (dbr^ Amplitude Figure Barker-13 sounding scattergram fo r Angle - Leicester path at 18:30 on 30/6/1997 at 4.92 MHz. 3.51,,-----, r C l 2 % 1.5 i UJ ol _L*_J ' j j - ^ i. : : ' ' j- r. * * \ q! Hour (UT) Figure Effective delay spread and Doppler spread for 30/6/

78 8 g 6 I 4 I I* I - e j & STANAG bps 10 > km ta Weak interferers Strong interferers <& ii i r? I $ 1 -»- * *«6e BFSK 600 bps Weak interferers Strong interferers I* I q>6 *6< ! »» Hour (UT) ft - - I * 9- <h 4 o e o-f 20 Figure Number o f evident interferers fo r 30/6/1997. Concluding remarks The examples given in this chapter show that over the course of a day it is common to find that different modem waveforms, that is waveforms using different modulation types and different data rates, having differing levels of performance with time. This suggests that there is a clear need for intelligent managing of HF data communications in order to allow the best use of the available channels to be made. This should be by the selection of the most appropriate modulation parameters for the given conditions. The performances of the waveforms that have been examined appear to vary depending on the basic modulation type and on the data rate employed. Having a variable SNR, which is often caused by a fading channel affects the PSK waveforms more so than the FSK type examined here. The nature of any background interference is a major factor in the performance of different modulation types. When a strong narrow band interferer is present this is far more likely to cause demodulation problems for PSK waveforms than for FSK waveforms. This is due to the high probability that the interferer is masking a section of the channel bandwidth that is vital for the correct reception of PSK. Conversely, with FSK (binary in this case) requiring 4-29

79 only two narrow segments of the channel bandwidth for signalling it is less likely that an interferer will affect reception. It was seen from the data collected on 29/6/1997 that a narrowband interferer caused the PSK waveform to perform poorly whilst the FSK waveform was largely unaffected. It is still, however, possible that the interferer and an FSK tone will be collocated, and in this case the FSK waveform will fail completely whereas a PSK waveform would probably still work, but at a reduced capacity. The number of propagation modes estimated from ionograms during this experimental period had little detectable effect on the performance of any of the waveforms examined. It is likely that the overall delay spread was minimal and as such, was not a problem. However, larger delay spreads are not uncommon and will be the cause problems for both PSK and FSK waveforms at other times. Using ionograms or monitoring the signal to noise ratio for particular channels will not necessarily allow the best waveform to be chosen for a given time as the information provided by these methods is not sufficient. If it were possible to monitor the channel conditions and determine from this information if another waveform might perform better, educated modulation changes could be made. The modulation could be changed in terms of data rate (by altering any forward error correction or by altering the signalling rate) or even by changing the signalling method (i.e. from FSK to PSK). Additionally, any such monitoring might give an early warning of when a particular waveform is about to fail and allow the communications system to switch to a more robust scheme instead of losing the link entirely. 4-30

80 Chapter 5 - Error rate derived link management The simplest comparison of performance for different waveforms is a measure of their bit error rate (BER) as a function of SNR over a simulated Gaussian noise channel. It is common for HF modems to support synchronous serial communications between the data terminal equipment (DTE) interfaces at each end of a link, this allows the true BER to be measured using the appropriate test equipment. The BER for several different waveforms was measured under Gaussian channel conditions using a pair of Fireberd 2000 bit error rate testers. To measure the BER the originating tester transmits a known pseudo random sequence of bits through the sending modem. At the receiving modem, the output bit stream is analysed to find how many bits of the known sequences have changed during transmission, the bit error rate is continually calculated for the duration of a test. The channel conditions were simulated using a DERA (NATO approved, CCIR report 549 compliant) HF channel simulator as described by Clarke and Davies [1995]. This was fitted in line between two HF modems as depicted in Figure 5.1 which is a schematic of the equipment used to measure the BERs. A selection of the BER results are presented below. Fireberd 2000 BER tester Fireberd 2000 BER tester Synchronous serial data HF modem HF baseband channel simulator HF modem Baseband audio Figure 5.1 Schematic o f equipment usedfor bit error rate testing. Illustrative performance data for various modulation schemes The widely used modulation type STANAG-4285, which is a 2-8-ary PSK scheme, is characterised for bit error rate performance at data rates bps in Figure 5.2, the results closely match those published in the STANAG document itself [NATO 1989]. It can be seen from this figure that STANAG-4285 will function reasonably (BER of 10-4 or 5-1

81 1.0CE better) at SNRs greater than -4 db. However, a SNR of at least 11 db is required for the highest illustrated data rate to function well. 1.00E E bps -* bps -* bps x 600 bps -*-1200 bps -*-2400 bps SNR (db) Figure Gaussian channel performance o f a typical STANAG-4285 modem fo r various data rates. An alternative way of looking at modem performance is to compare differing modulation types running at the same data rate. Figure 5.3 illustrates the performance over a Gaussian noise channel of three different modulation schemes at 600 bps. The STANAG-4285 and MIL-STD A waveforms are both 2-8-ary PSK, but have different formats for data encapsulation and bit/frame synchronisation, hence their similar performance. The FSK waveform performs marginally better than the PSK types below -1 db SNR, at all other points its performance is markedly poorer. It is common to find multiple waveforms available in modem modems and as a result, this could allow the communicator a wide choice for his communications needs. Figure 5.4 is a performance comparison for a range of waveforms and data rates that will typically be available in a current modem. Here three waveform standards are illustrated: MIL-STD A 39 parallel tone (1200 bps), STANAG ary PSK ( bps) and MIL-STD B QPSK (3200 bps), 8-ary PSK (4800 bps) 16-QAM (6400 bps), 32- QAM (8000 bps) and 64-QAM (9600 bps). The evaluation of the MIL-STD B waveforms was undertaken using one o f the procedures and the same BER and simulation 5-2

82 equipment as described by Gillespie and Trinder [2000]. The results of this evaluation are in close agreement with those in the modem standard document [US DoD, 2000] and those presented in Nieto [2000] for a commercially available modem. 1.00E E E E-03 -STANAG MIL-STD A -FSK 1.00E E E SNR (db) Figure Gaussian channel performance o f three typical modulation schemes at a data rate o f 600 bps. 1.00E E E-02 «1.00E-03 m 1.00E E-05 MIL-STD A 39 tone 1200 bps -m - STANAG bps STANAG bps STANAG bps * STANAG bps STANAG bps f - STANAG bps MIL-STD B 3200 bps MIL-STD B 4800 bps MIL-STD B 6400 bps -» MIL-STD B 8000 bps MIL-STD B 9600 bps 1.00E E SNR (db) Figure Gaussian channel performance o f various typical modulations schemes implemented in a modern modem fo r various data rates. 5-3

83 How error rate derived management could be achieved If a measure of a link s recent error rate is available, this can be used to help decide whether it would be advantageous to change any modulation parameters for future transmissions. With modem modulation types, such as STANAG-4285, MIL-STD A serial tone and MIL-STD OB, it is common for the modem to allow user control of the on-air data rate and the interleaver length used. Both of these affect the performance of the modem in terms of data throughput rate and error rate when used over real HF channels. Bit interleaving is often used in HF modems to distribute the transmit data bits in order to minimise the effects of error bursts on any forward error correction employed as the forward error correction techniques used are more effective against random errors rather than burst errors. For a simple system using only one modulation type, based on Gaussian noise channel BER against SNR curves such as in Figure 5.2 and throughput against SNR curves as in Figure 5.5, it is possible to imagine a system that would change the data rate of the modem as the channel conditions improve or decline in order to maintain the best throughput possible. The throughput against SNR curves can be used to calculate the optimal SNR at which to switch data rates, this can then be translated to the expected BER. The optimal point to change data rate is where, in the case of changing up a data rate, the new data rate starts performing significantly better than the current rate. For example, with STANAG and changing data rate from 1200 bps upwards, this would be best undertaken at an SNR of around 5 db (corresponding BER of around 10 5) to show a useful improvement in data throughput. The above theory makes the assumption that changing from a data rate which is performing with zero, or very few errors, to one which has the net result of providing a much greater throughput yet has a significant number of errors is acceptable to the higher level protocols. However, with modem frame based protocols this is not the case. For example, if an HF link were to be used as an extension to a TCP/IP network, it would be common to use transmit data block sizes (maximum transmittable unit or MTU) of around 1500 bytes. For TCP to work efficiently, these blocks must arrive mostly error-free even though this protocol supports retransmissions. This is because retransmissions significantly affect the performance of this protocol due to two reasons, firstly, the whole frame must be retransmitted, and secondly, TCP inserts an exponentially increasing delay between 5-4

84 retransmissions in an attempt to counteract link congestion (i.e. it treats frames lost due to errors as if they were lost due to congestion). Therefore, over a link with a high BER, TCP/IP will not work well I 1800 n I C O) bps -150 bps -300 bps 600 bps bps bps SNR (db) Figure Calculated throughput performance over a Gaussian channel fo r a typical STANAG modem at various data rates. The probability of having to retransmit a frame of data can be calculated given the bit error rate and the frame length. The probability of a bit being received correctly (Pc) is given by Equation 5.1 where Pe is the probability of a bit being received in error. Pc = l -P e(5.1) The probability of a block of «-bits received ( Pbc) being received correctly with no errors is given by Equation 5.2, and finally the probability of a block of /7-bits being received with at least one error is given by Equation 5.3. pbc= p ; (5.2) Pbe=\ -P hc(5.3) Therefore with a BER of 10-4, it would be expected that approximately 70% of 1500 byte TCP/IP frames would need retransmitting due to errors. If the BER was lower, say 10'5, then this figure falls to about 11% of the frames. 5-5

85 Realistically, it is more usual to encapsulate and fragment wired network protocols in robust, data link layer protocol frames such as with STANAG-5066, and for the data link layer protocol frames to use block sizes of around 200 bytes. For the two BERs quoted above, the frame error rate for these smaller frames would be approximately 15% and 1.6% respectively. In practice, Gaussian noise channel characteristics are not found over sky wave HF links and as a result, this has an effect on the bit error distribution. Under Gaussian noise conditions, the BER can readily be used to estimate the frame error rate, this is useful as described above where frame based protocols rely on the retransmission of whole frames when errored. Under non-gaussian noise conditions as experienced on HF links, the bit errors become clustered, and the frame error rate is actually lower for a given BER than might be expected. Based on the BER and throughput curves for STANAG-4285 introduced above, a simple modulation parameter change, or in this case, a data rate change (DRC) technique can be formulated. A possible scheme would be: at a suitable time during the communications over a link, for example when acknowledgements are sent for frames of data received, if the error rate is too high, a command to change the data rate should be sent from the acknowledging station. If this command is accepted and itself acknowledged, the data rate at both ends of the link should then be changed. Then, a further acknowledgement should be sent to be sure that that data rate change has been successful and that both ends are synchronised and ready to commence data communications again. The BER value should be chosen as being a point where changing data rate upwards will result in the new data rate performing adequately and with a greater throughput. DERA data rate change test bed A test bed, originally built by DERA to evaluate the performance of the (then) proposed STANAG-5066, has been further developed to enable data rate change mechanisms to be evaluated. A block diagram of the test bed can be found in Figure

86 291 km sky wave path N / HF transceiver HF transceiver PSTN modem HF modem PSTN modem HF modem Portsdown West site Angle site Figure Schematic o f the Data Rate Change test bed. The DRC programs running on the PCs controlled the HF modems and radio systems via serial connections allowing the modulation parameters of the modems and the transmit / receive frequencies to be changed automatically. The DRC program worked in either transmit or receive mode, this dictating whether the site was sending or listening for data. The DRC test bed operated according to a time schedule whereby the transmit program generated and sent blocks of data, which were protected by checksums, across the HF link where they were received and any errors recorded by the receive program. The transmission period lasted for two minutes, after which an acknowledgement message was sent from the receiving side, which could optionally contain a request to increase or decrease the data rate, and was known as a DRC request. The decision whether to send a DRC request was made by the DRC algorithm of the receiving program based on the frame error rate. If no DRC request was made, then a three way handshake was used to ensure both ends were synchronised (Figure 5.7) before the transmitter continued and sent another two minute transmission. The two way handshake took approximately 20 seconds to complete. However, if a DRC request was made, a five way handshake process was used to negotiate the data rate change and to ensure that both ends of the link were operating with correctly agreed parameters, see Figure 5.8. This five way handshake and subsequent changing of parameters took approximately 40 seconds. If the handshake or DRC process failed, an error recovery mode was entered and maintained until both sides became 5-7

87 synchronised once more and a common data rate was used to resume the data block transmission. RX TX 1 Transmission ACK 2 A ACK confirm 3 Ready for data Data transmission recommences Figure Data rate change test bed three-way handshake. RX TX 1 DRC REQ ^ 2 A DRC REQ ACK 3 DRC ACK Modem data rate change occurs 4 A New data rate ACK 3 ACK confirm Data transmission recommences Figure 5.8- Data rate change test bed five way handshake. The test bed software was designed to allow different algorithms for the decision processes to be readily implemented and tested on air (a new algorithm could be added by modifying C language routines). Not only could the data rate change algorithm be modified, but also the methods for error recovery and subsequent resumption of data transfer. Results of error rate derived management experiments The algorithm discussed here that was implemented on the DRC test bed was based on one suggested in FED-STD-1052 [National Communications System, 1995], this proposed the simple technique of halving the data rate when the frame error rate became greater than 5-8

88 50%, until the minimum data rate was reached, and doubling the data rate when the frame error rate was zero until the maximum data rate was reached. Further analysis of data rate change techniques has been described by Trinder and Brown [1999] where it was shown that this algorithm is optimal for Gaussian noise channels and STANAG The results in the form of data rate selected by the algorithm (and successfully switched to) against time of day for one DRC test are shown in Figure 5.9. This experiment ran from 11:15 hours on 30/10/1998 until 16:08 hours on 31/10/1998 at a frequency around 4 MHz over the Angle to Portsdown West path. A clear distinction between day and night time performances can be seen with only the lowest data rates being used between 19:00 and 08:30 hours approximately I 1200 re :00 12:00 14:00 16:00 18:00 20:00 22:00 00:00 02:00 04:00 06:00 08:00 10:00 12:00 14:00 16:00 18:00 Time (UT) Figure D ata rate change algorithm (DRC) data rate selected results (30-31/10/1998). Another example of the data rates achieved when using the DRC algorithm is presented in Figure These data were collected over a 25 hour period between 25-26/10/1998, again on a frequency around 4 MHz. Similarly to the example above, there is a clear day / night transition with lower data rates being used during the hours of darkness. The switch to lower data rates appears to take place at around 15:30 hours, and back to higher data rates at around 07:00 hours. The main difference between these results and those of Figure 5.9 is that in this case, there are quite a few oscillations between the two highest 5-9

89 data rates (i.e and 2400 bps). As each change in data rate takes an extra 20 seconds over the normal handshaking process, it is possible that the algorithm was wasting transmit time in striving for the optimal data rate. It is worth noting that the use of a single frequency for day and night time communications over a fixed link is not optimal due to the changing MUF. From the results presented, the 4 MHz frequency was clearly a good choice for the daytime due to the high data rates being supported. Over the Angle to Portsdown West path a lower frequency may have been of benefit during the night time hours, however, in this case a lower frequency channel without excessive noise was not available. An automated frequency management system may have be able to make a better choice of channel and resulted in better night time performance : :00 18:00 20:00 22:00 00:00 02:00 04:00 06:00 08:00 10:00 12:00 14:00 Time (UT) Figure D ata rate change algorithm (DRC) data rate selected results (25-26/10/1998). Whilst the results of the DRC tests show the system successfully changing data rates across the course of the day with lower rates at night, which might be expected, they do not indicate whether performing these data rate changes is beneficial to a communicator. To establish whether this is so, it is necessary to compare the data throughput achieved using the DRC system with a predicted throughput that would have been achieved without the DRC process, i.e. data rates selected prior to communications taking place. To do this, 5-10

90 several scenarios of fixed data rates need to be examined as alternatives. For comparison purposes, scenarios including fixed data rates of 75, 150, 300, 600, 1200 and 2400 bps are examined here. The simplest method of estimating data throughput for a pre-selected data rate system is to assume that error-free transmissions are possible throughout the day and night. For such a system it is also assumed, for the purposes of these calculations, that a three way handshake as shown in Figure 5.7 would still be used to maintain synchronisation between transmitting and receiving systems and hence the time taken to perform this handshake will be used in estimating the throughput figure. This handshake process takes approximately 20 seconds with the DRC system. The estimated throughput for a given fixed data rate using the three-way handshake of the DRC process, assuming 100% success, can be calculated using Equation 5.4 below. Where texperimenl is the duration of the experiment in seconds, ttransmisslo is the duration of a single transmission period in seconds (including handshaking), n is the number of frames sent per transmission and s is the size in bytes of each frame. The transmission period employed by the DRC system was 2 minutes, the frame size was set to 200 bytes and the number of frames sent was dependent on data rate. Throughput (bytes) = e**>enment xnxs (5.4) t transmission The throughput achieved using the DRC process was found by summing the number of payload bytes received without error over the duration of the test. For the results presented in Figures 5.9 and 5.10 from 30-31/10/1998 and 25-26/10/1998, these totals were just over 6 million and 3 million bytes respectively. For fixed data rates of bps and for the case of using 75 bps during the night and 600 bps during the day, the maximum possible data throughput values are tabulated in Table 5.1 along with the percentage throughput gain that would be achieved by using the DRC system instead. The results show that the DRC system operating over a real HF channel offers significant gains over a theoretical fixed data rate system operating at data rates up to 600 bps for the first test period and up to 300 bps for the second test period even in the unlikely scenario of error-free conditions. 5-11

91 Fixed data rate (bps) Percentage increase in throughput achieved using DRC on 30-31/10/1998 compared with fixed rate system. Percentage increase in throughput achieved using DRC on 25-26/10/1998 compared with fixed rate system % 588% % 244% % 72% % -11% 1,200-19% -54% 2,400-60% -77% 75 night / 600 day 164% 94% Table A comparison o f throughput gain achieved using a DRC system and an error-free fixed data rate system fo r various data rates. For the two periods which DRC results are presented above (30-31/10/1998 and 25-26/10/1998), it is clear that data rates greater than 75 bps were only supported at a reasonable error rate for a fraction of the time. Therefore, the comparison results in Table 5.1 for the theoretical error-free fixed data rate system are misleading. To undertake a more realistic comparison that does not assume error free conditions and continuous support, the duration for which each data rate was supported by the channel was taken from the DRC data and tabulated in Table 5.2. for both periods. The error rate measured by the DRC system was used in conjunction with the supported durations to form an estimate of the overall data throughput for each fixed data rate. The results of these estimates are tabulated in Table 5.3 in the same format as previously. The more realistic results in Table 5.3 indicate that the DRC system achieved a useful throughput gain over any of the fixed data rate scenarios when taking in to account the duration of channel support and the measured error rate for each data rate. 5-12

92 Data rate Duration data rate supported over 30-31/10/1998 period (minutes) Duration data rate supported over 25-26/10/1998 period (minutes) Table 5.2- Durations o f channel support fo r given fixed data rates based on DRC results. Fixed data rate (bps) Percentage increase in throughput achieved using DRC on 30-31/10/1998 compared with realistic fixed rate system. Percentage increase in throughput achieved using DRC on 25-26/10/1998 compared with realistic fixed rate system % 938% % 715% % 321% % 139% 1,200 64% 37% 2,400 17% 20% 75 night / 600 day 154% 109% Table A comparison o f throughput gain achieved using a DRC system and a realistic fixed data rate system fo r various data rates From the DRC throughput results, it is clear that the test performed on 25-26/10/1997 had a lower throughput relative to that performed on 30-31/10/1997 (around 3 million and 6 million bytes respectively). Further examination of Figure 5.10 reveals a potential problem where the data rate oscillates between the two highest rates and an analysis of the effect of this is presented here. During the course of this experiment, there were a total of 311 transmission periods where data was sent across the link. The frame error rate was analysed after each period and a request to change data rate was made if the algorithm dictated it was necessary. There were 5-13

93 6 instances of failed data rate change handshakes when attempting to increase data rate from 1200 to 2400 bps, these were due to either, an interference signal being present, or the channel conditions being good enough to support 1200 bps well, but not satisfactory for 2400 bps. In total, there were 24 data rate changes made as a result of oscillations from 1200 to 2400 and back to 1200 bps (oscillations only counted if completed within a ten minute period) including the failed handshakes. These data rate changes will have taken 20 seconds each over and above the normal handshake and this amounts to 480 extra seconds altogether. As this 480 seconds is over a 25 hour test period, only a small amount of time was consumed in oscillating between the two highest data rates. It is possible, under other circumstance such as if the channel never became quite sufficient for 2400 bps, that these kind of oscillations would waste a much larger period of time. Concluding remarks Using the data rate change algorithm tested above results in a greater throughput across a 24 hour day / night period compared to using a fixed data rate if there are significant changes between daytime and night-time propagation characteristics. This implies that the communicator can make better use of the HF allocations available, and can reduce the effort put into managing links if an automated data rate change system were to be employed. Greater data throughput on a link can lead to shorter transmission times thus lowering the probability of detection. Alternatively, the link could be better utilised and more data could be passed over it, potentially reducing the traffic on other communications systems. It should be relatively simple to integrate a data rate change mechanism into modem two way, point-to-point, communications systems. Protocols such as STANAG-5066 can already accommodate the addition of a data rate change mechanism. However, care must be taken when recommending any system that could delay the transmission of critical messages. The communications system might not always have a suitable modulation scheme to change to at a given error rate, so some degree of granularity has to be accepted with current waveforms. This means that changing from one modulation scheme to another (possibly to increase the data rate) may involve changing from an error-free condition to one where errors are inherent. 5-14

94 Using frame error rate statistics alone to make decisions on modulation parameter changes is one of the simplest techniques, but it is not the best. A greater understanding of the underlying HF channel will allow more accurate decisions to be made based on the changing propagation and interference conditions. Other methods of channel evaluation could provide useful inputs to modulation parameter change algorithms and are likely to be of great benefit in reducing oscillatory behaviour such as that seen in the second set of DRC results presented in this chapter. 5-15

95 Chapter 6 - Ionospheric probing to aid modem waveform selection In the previous chapter it was suggested that better use could be made of an HF channel if the modulation parameters of the modem were adjusted according to the prevailing channel conditions. The use of error statistics available from communications protocols provides one method of monitoring the state of the channel. However, this is an indirect method and is one that cannot be used reliably to recommend alternative modulations based on anything but simple Gaussian noise channel models of modem performance. Real HF skywave channels exhibit conditions that are often very different from Gaussian noise channels due to the effects noted in Chapter 1, namely multipath delay, fading and interference etc. These conditions need to be identified and monitored if a more sophisticated system for modulation parameter adjustment is to be developed. Additionally, the performance of available modulation schemes needs to be known for realistic varying conditions that might be expected over HF paths. Data modulation performance characteristics The many different modulation types that are commonly available to a communicator have vastly differing performance characteristics when used over HF paths. In addition to the basic modulation types having different demands of the channel, most waveforms support variable, user selectable, parameters such as data rate etc. all of which further modify the characteristics and requirements of the waveforms. It has already been seen in Chapter 5 that different modulation types can be characterised for Gaussian noise channels, a technique for performing a fuller characterisation is presented here. This includes the ability to alter the Doppler and delay functions of the simulated channel in addition to that of the SNR value. Multi-dimensional characterisation of modem performance Arthur and Maundrell [1997] describe a system to characterise HF modems in terms of Doppler spread, delay spread and SNR for a given BER. This system employs a technique of measuring the BER for each combination in a range of Doppler and delay spread (conditions generated by an HF channel simulator) for various values of SNR. The results 6-1

96 of this characterisation are usually presented as a graphical representation of a 3- dimensional surface. Example results of such a characterisation are presented in Figure 6.1 which was produced using the technique and equipment described in the paper. This plot shows a surface indicating the SNR required for a typical STANAG bps modem to achieve a BER of 10 3 given the delay and Doppler spread conditions of the x and y axes. CO a. z to Multipath (ms) Figure dimensional characterisation p lot fo r a typical 1200 bps STANAG-4285 implementation. If the values that make up such a plot were to be gathered for every modulation type that a communicator had available and if the appropriate channel parameters were measured then it would be possible to select the best performing modulation type available for the current channel conditions. 6-2

97 Ionospheric conditions to be expected Modulation schemes are not designed to cope with excessive delay or Doppler spreads as these are limited in all current practical communications. That is to say, there may be many propagation modes present, but not all are detectable and those that are not are likely to be insignificant to the receiving modem. The multi-dimensional surface plot of a waveform s performance characteristics shows that both the Doppler and delay spread parameters can have severe effects on the BER at their extremes. However, with a knowledge of the intended HF path, it is feasible to almost rule out the possibility of one of these two parameters extremes being reached. Propagation characteristics differ between high, mid and equatorial latitude paths due to the differing nature of the interaction of the ionosphere with the solar-terrestrial environment in these regions. High and equatorial latitude paths tend to display larger frequency dispersions than mid-latitude paths. Davies [1990] explains these differences in detail. These high Doppler spreads make radio communication difficult and reduce the speed of achievable data rates as the modulation schemes used compensate by using increased coding or more robust signalling. Warrington et al [2000] describe a campaign where Doppler spread conditions for various frequencies over two high latitude paths were observed during a 17 day period. The results of this indicate that Doppler spreads of 5 Hz or greater are not uncommon (typically over 7% of the time during the campaign) at these latitudes. Over mid latitude paths, Doppler effects are minimal compared to those at high latitudes and delay spread is far more likely to be the more important factor in HF communications resulting in a limitation of the maximum signalling rate possible. In order to determine how large the delay spread is likely to extend to, a few simple calculations were used to produce figures for the path delay imposed by communicating via different propagation modes. It is reasonable to assume that the majority of mid latitude HF communications do not experience more than five hops in propagation. Based on this premise, the maximum delay spread that is likely to be encountered is limited to less than 10 ms, assuming ongreat circle propagation. To demonstrate this, two plots are presented in Figure 6.2. The upper plot shows the approximate path delay for a simple 1 hop E mode and that for a more complex, 5 hop F2 mode based on a flat earth model. The lower plot is a curve indicating the maximum delay spread that would be resultant if both the 1 hop E and 6-3

98 5 hop F2 propagation modes were supported. From this second plot, the delay spread indicated is around 9 ms at very short path lengths and reduces to around 4 ms at path lengths of around 3000 km. It is appropriate to use a flat earth model for this particular calculation as the delay spread estimated is due to the path delay differences between the two propagation modes, and this is at its greatest at short transmitter to receiver distances. Based on these results it is reasonable for a channel evaluation or modulation selection system to have a delay spread range of up to about 10 ms to match expected conditions. 15 Path delay for 1 hop E and 5 hop F2 modes _ 10 1 hop E 5 hop F Delay spread for propagation over 1 hop E and 5 hop F2 modes I Distance between TX and RX (km) Figure Propagation modes required to cause varying delay spreads. Modulation types available for differing channel parameters Given the expected ranges for delay and Doppler spreads likely to be encountered over high latitude and mid latitude paths, and given the kind of performance information that the multi-dimensional modem characterisation plots yield, it is possible to list appropriate modulation schemes for varying channel conditions. As an example, for the STANAG- 4285, MIL-STD B and FSK modulation schemes, Figure 6.3 illustrates which data rates and interleaver settings (for a selection of rates and interleavers) work at delay spreads of up to around 6 ms. The slower data rates of the modem waveforms can cope with larger delay spreads, whereas the faster rates demand smaller delay spreads. The tolerance to multipath delay o f the FSK waveforms is limited by the signalling technique 6-4

99 and is quite poor at rates greater than 75 bps. A similar chart is presented in Figure 6.4 for Doppler spreads of up to around 40 Hz, here both the data rate and the interleaver length (where applicable) affect the tolerance of Doppler conditions. Longer interleavers and lower data rates work at larger Doppler shifts than shorter interleavers and higher data rates. Here it can be seen that the slower FSK rate exhibits exceptional tolerance to Doppler spreads. From these charts it is possible to find a subset of the waveforms that are appropriate for high latitude communications. Based on the figure for Doppler spread experienced given earlier, only a small subset of the modulation schemes in Figure 6.4 would be suitable. For mid-latitudes, Doppler spread must still be taken into account if the higher data rate waveforms are to be used and multipath spread will be critical if it approaches 4 ms. MIL-STD B 9600 VS MIL-STD B 9600 VL MIL-STD B 8000 VS MIL-STD B 8000 VL MIL-STD B 6400 VS MIL-STD B 6400 VL MIL-STD B 4800 VS MIL-STD B 4800 VL _ MIL-STD B 3200 VS E MIL-STD B 3200 VL STANAG S S a STANAG L STANAG S STANAG L STANAG S STANAG L STANAG S STANAG L FSK 600 FSK 300 FSK Maximum delay spread (ms) F igure W orking delay spread regions fo r various modulation schemes. 6-5

100 MIL-STD B 9600 VS MIL-STD B 9600 VL MIL-STD B 8000 VS MIL-STD B 8000 VL Ml L-STD B 6400 VS MIL-STD B 6400 VL MIL-STD B 4800 VS MIL-STD B 4800 VL _ MIL-STD B 3200 VS» f ; MIL-STD B 3200 VL STANAG S g Q STANAG L STANAG S STANAG L STANAG S STANAG L STANAG S STANAG L FSK 600 FSK 300 FSK Maximum Doppler spread (Hz) Figure Working Doppler spread regions fo r various modulations schemes. The need to measure or estimate channel parameters to aid modulation-type selection Given that the requirements of the various modulation schemes can be established using techniques such as the multi-dimensional characterisation, as described earlier in this chapter, it is necessary to have a good measure of the current channel parameters in order to select the best scheme to use for communications. Using ionospheric prediction programs such as the IONCAP system introduced in Chapter 2, it is possible to estimate the modal content to be expected over a given path with knowledge of the frequency of operation, details of the antennas and time of day. This can be used to further estimate the delay spread. Knowledge of the path geometry may allow estimates to be made of the Doppler spreads likely to be encountered. With these estimates, a modulation scheme that is likely to work yet is still efficient can be chosen. One method of actually determining the current propagation conditions is to use an ionospheric probe such as those introduced in Chapter 2. These include techniques such as the chirp-sounder ionosonde and the DAMSON systems. Probes such as these can reveal details about the propagation path which have an effect on the performance of data modulation schemes. 6-6

101 Directly probing the ionosphere with a pulse compression waveform, as in the DAMSON system, can be used to derive the channel scattering function using only a moderate amount of transmit power. The scattering function when plotted shows how the signal is spread by the channel in the time and frequency domains. This technique works by sending a known binary sequence using bi-phase shift key modulation over the HF channel, with the received signal being passed through a matched filter to calculate the cross-correlation function. The result is a measure of the channel impulse response. At each lag of the crosscorrelation the output is operated on by a fast Fourier transform routine to determine the frequency domain power spectrum. Using the Barker-13 sequence as a probe As introduced in Chapter 2, the Barker sequences have the special property that for all lags other than zero, the autocorrelation value can only be equal to 0,1, or -1. The length of the sequence determines which of these are present. At the zeroth lag the autocorrelation has a value equal to the code length itself. As an example, the autocorrelation function for the 13 bit Barker sequence is shown in Figure 6.5 where the peak autocorrelation value of 13 is clear. The only other values that the autocorrelation of this particular sequence yields are 0 and 1, thus the Barker criteria is met. Rao and Deshpande [1988] explained that the patterns that fulfil the Barker criteria are limited in maximum length to 13 bits. However, the 13-bit Barker code turns out to be very useful for use as an ionospheric channel probe. If this code were to be transmitted repeatedly at 2400 baud, with no gap between one code finishing and the next starting, it would allow multiple, time delayed, modes to be resolved at up to approximately 5.4 ms apart. This is due to the duration of a single code sequence. The length of this 13 bit code can theoretically result in a processing gain of just over 22 db. This gain eases the signal detection problem in the presence of channel noise. As described earlier, the DAMSON system employs the Barker-13 code to measure the channel scattering function. In this system these sequences are not transmitted back to back as just mentioned, but following a Barker-13 code sequence a period of silence is sent (i.e. the transmitter is switched off). This is in order to increase the length of the delay spread that is unambiguously measurable. The Barker code lasts for just over 5.4 milliseconds and the silence period is a further 17 symbols in length making the duration of one coded pulse and the following silence period equal to 12.5 milliseconds, 6-7

102 see Figure 6.6. Therefore two propagation modes can be distinguished, given suitable power in each, at up to 12.5 ms apart. In the frequency domain, the Doppler range that can be unambiguously measured with the DAMSON system is +/- 40 Hz from the nominal sub-carrier frequency. The limiting factor here is the length of the code sequence (frequency range in hertz is equal to the number of codes transmitted per second). This frequency range was required for the high latitude research work that the DAMSON system was built for <u 8 I c = V W W n ^v v x a a / Lag Figure B arker-13 autocorrelation function. Barker-13 sequences (13 symbols) 12.5 milliseconds Silence periods (17 symbols) Figure 6.6- Schematic o f how the 13-bit Barker code is used in DAMSON. 6-8

103 A probe such as the DAMSON waveform described here is used in place of any data communications waveform. Therefore probing in this manner will cause data transmission time to be lost, and may result in delayed messages and a lower overall data throughput. Results from on-air Barker-13 probe experiments A series of experiments were performed over the Angle - Leicester path in order to compare soundings made using a Barker-13 probe sequence with those from a FMCW ionosonde. These two methods differ considerably in operation but should yield matching results for the measurements of the multipath spread and the number of propagation modes detected. Additionally, each method can reveal extra channel information that could also be of value to the communicator. The equipment used to perform these experiments has been described previously in Chapter 3. The configuration was such that the Barker-13 waveform was transmitted for 10 seconds followed shortly by a STANAG-4285 signal for a further 150 seconds. This complete sequence was sent every 30 minutes. Additionally, ionograms were transmitted and collected every 5 minutes for the same path. The first comparison of a Barker-13 probe and an ionogram was collected at 01:30 hours on 14/6/1997. The ionogram is presented in Figure 6.7 with the vertical red line indicating the spot frequency on which the Barker-13 probe was transmitted. The three black traces represent the observed propagation modes (frequencies and time delays associated with propagation), the straightest being caused by E-region propagation and the two curved traces from F-region propagation. As the red line does not intersect any of the traces of the ionogram, there is no propagation between the transmitter and the receiver at that particular frequency. Figure 6.8 is from the analysis of the Barker-13 probe after it has been transmitted over the HF channel. The absence of a clear peak in the amplitude / delay pane and of a clear peak in the frequency / amplitude pane indicates that the probe sequence was not detected. 6-9

104 5 A n g le 1 4 /0 6 / :3 0 U T ( ) *3.5 E ^ 3 GJ <D > TO * <D a 1.5 J, I V, M H z 4A ' 'I, *! 1 ^ V'<J Figure Ionogram collected at 01:30 on 14/6/1997fo r Angle - Leicester path. Barker-13 sounding: MHz 14/06/1997 1:30 UT Amplitude Figure Barker-13 sounding scattergram for Angle - Leicester path at 01:30 on 14/6/1997 at 4.92 MHz

105 Another example was collected later on the same day at 10:00 hours. The ionogram in Figure 6.9 shows a clear intersection of the red frequency marker line and the black propagation trace, therefore at the time the ionosonde swept through the indicated spot frequency there was propagation between the transmitter and the receiver. This propagation appears to be by means of an E-region reflection due to the appearance of the ionogram trace. The results of the Barker-13 probe are displayed as a scattergram plot in Figure The upper pane of this plot shows a reasonable peak in amplitude and indicates that the Barker-13 sequence has been detected amongst a fairly high level of background noise. The centre pane of the plot indicates the relative power of the detected signal against time and frequency axes. This scattergram shows no evidence of any significant secondary propagation modes. 5 A n g le 1 4 /0 6 / :0 0 U T (1 2 0 ) E 3 CD M H z Figure Ionogram collected at 10:00 on 14/6/1997fo r Angle - Leicester path. 6-11

106 Barker-13 sounding: MHz 14/06/ :00 UT Delay (ms) Amplitude Figure Barker-13 sounding scattergram fo r Angle - Leicester path at 10:00 on 14/6/1997 at 4.92 MHz. To demonstrate the effects of multiple propagation modes further examples are presented in Figures 6.11 and The data for these plots was collected at 19:00 hours on the same day as the previous examples presented here. The ionogram for this time shows strong E- region and slightly weaker F-region propagation around the frequency of interest. The time delay between the two propagation modes can be measured from the ionogram to be about 1.3 ms. The distance between the peaks along the time axis on the Barker-13 sounding plot is also 1.3 ms

107 A n g le 1 4 /0 6 / :0 0 U T ( ) * M H z Figure Ionogram collected at 19:00 on 14/6/1997fo r Angle - Leicester path. Barker-13 sounding: MHz 14/06/ :00 UT R e c e ive d. p.owe r Id B.r) Amplitude Figure Barker-13 sounding scattergram for Angle - Leicester path at 19:00 on 14/6/1997 at 4.92 MHz. 6-13

108 STANAG-4285 equaliser training sequence The STANAG-4285 modulation scheme uses a frame structure in order to carry its data payload bits over the HF channel. Figure 6.13 is a simple illustration of the STANAG frame format. A frame consists of 256 symbols. The first 80 of these are used for frame synchronisation, 48 others are for reference during the reception / decoding of a frame and the remaining 128 symbols are the data symbols. The modulation rate for all data rates is 2400 baud. Data symbols Synchronisation symbols Reference symbols Figure Schematic o f STANAG -4285fram e structure. The synchronisation symbols are intended for use by the receiving modem in order to allow accurate detection of the signal, for the correction of sub-carrier frequency offsets between the transmit and receive systems and for equaliser training / channel estimation. The format of the 80 symbol synchronisation block is made up of two full 31 bit pseudorandom noise (PN) sequences and the first 18 bits of another, its format is shown in Figure bits 31 bits 18 bits Figure STANAG-4285 synchronisation sequence form at. These 31 bit PN sequences lend themselves readily to working as channel probes in the same way that the Barker-13 sequence described above does. The autocorrelation function of this PN sequence (Figure 6.15) is not quite as good as that of the Barker codes, but it is certainly useable for channel evaluation. This sequence when transmitted as part o f the 6-14

109 synchronisation block in STANAG-4285 can allow propagation modes up to ms apart to be resolved. In the frequency domain, the Doppler range is +/ Hz Lag Figure bit P N sequence (from STANAG-4285 training sequence) autocorrelation function. This sequence that is embedded in a standard modulation type is useful as a channel probe. It has the advantage over the Barker-13 probe method, described earlier in this chapter, in that data communications do not need to be stopped to make use of it. It is therefore possible to detect changes in channel conditions whilst data is being exchanged, this information may be used to aid decisions on changing modulation parameters. Example scattergram An example scattergram produced by cross-correlating the known 31 bit PN sequence with a sampled STANAG-4285 transmission is presented in Figure 6.16 below. This data set was collected at 09:00 on 15/7/1997. At around the same time an ionogram for the same path was recorded and from Figure 6.17 can be seen to detect two propagation modes at the frequency of interest (highlighted in red). The scattergram plot shows one large amplitude peak in both the frequency and time domains with one other significant peak in 6-15

110 the time domain. Other undulations are present in the time domain, but rather than indicating further transmission modes, these are artefacts o f the PN sequence. Stanag-4285 sounding: MHz 15437/1997 9:00 UT Received power (dbrl Amplitude Figure STANAG-4285 sounding scattergram fo r Angle - Leicester path at 09:00 on 15/7/1997 at 4.92 MHz. To further examine this approach to gaining channel information, a Barker-13 signal was transmitted on the same frequency immediately prior to the STANAG-4285 modem signal. This probe signal was sampled and its resulting scattergram is presented in Figure The two propagation modes detected in the other results are also clearly visible using this method. 6-16

111 A n g le 1 5 /0 7 / : 0 0 U T ( ) CD * M H z Figure Ionogram collected at 09:00 on 15/7/1997fo r Angle - Leicester path. Barker-13 sounding: MHz 15/07/1997 9:00 UT Amplitude Figure Barker-13 sounding scattergram for Angle - Leicester path at 09:00 on 15/7/1997 at 4.92 MHz. 6-17

112 Limitations of using the STANAG-4285 synchronisation sequence for channel evaluation The synchronisation sequence of STANAG-4285 as discussed above can be used to measure some useful channel parameters, these can be used to help make decisions about when to change modulation type and what modulation type to change to. The main limitation of using this technique for channel evaluation compared with using the Barker- 13 waveform is the reduced Doppler range due to the code repetition rate. Examples of channel evaluation parameters and measured modem performance over the Angle - Leicester path It has been shown that pulse compression probes such as the Barker-13 type mentioned earlier can provide accurate measurements of multipath and Doppler spread. To illustrate the effects of these conditions on data waveforms, another period of transmissions was undertaken. The procedure for this set was to transmit modem signals in bursts interspersed with transmissions of the Barker-13 probe signal to allow a comparison between the modem performance and the channel evaluation results. The first example was collected on 14/6/1997. Figure 6.19 is a plot of average data throughput against time of day for STANAG bps and 2400 bps waveforms and shows the lower speed waveform outperforming the higher speed waveform. The effective multipath delay and Doppler measurements are presented in Figure 6.20 and an estimate of the number of weak and strong interferers in Figure In Figure 6.22 the results of a Barker-13 probe are presented and show a measure of the instantaneous channel conditions at around 11:00 hours. At this time, according to the throughput plot, the performance of both modem waveforms dropped considerably. The probe results indicate that although the signal was detectable and that two equal power modes were resolvable with a short delay spread, the signal to noise ratio was quite poor. The SNR was estimated to be around db using the method described in Chapter 4 (ESNIR). Although the SNR was poor at this time, it is more likely that the Doppler spread was responsible for the almost total failure of the waveforms. Figure 6.23 is a magnified version of the Barker-13 scattergram and allows the Doppler spread to seen more clearly. At this time the Doppler spread approached the point where the 2400 bps waveform, if not the 1200 bps waveform, fails to function (around 2 Hz and 6 Hz respectively). Figure 6.24 shows a STANAG

113 scattergram taken around the same time, this too indicates a reasonably large Doppler spread. As the day progressed, the SNR became better and the 1200 bps waveform performed adequately, although at around 20:30 hours another dip in the performance of both modems is noticeable from the throughput curves. A Barker-13 scattergram using data from this time with a zoomed frequency axis is presented in Figure This shows again a large enough Doppler spread to cause demodulation difficulties for the 2400 bps waveform. 150 STANAG bps o STANAG bps S Hour (UT) Figure Average throughput fo r STANAG and 2400 bps on 14/6/1997 over Angle - Leicester p a th. 6-19

114 I 0.5 m «3 Q. _ Hour (UT) Figure Effective delay spread and Doppler spreadfor 14/6/1997. STANAG bps Weak interferers Strong interferers 4 I* -»<» fy *» * -J>O G O a STANAG bps Weak interferers Strong interferers 2 4 cj) logtw «: ;Y-« Hour (UT) Figure Number o f evident interferers for 14/6/

115 Barker-13 sounding: MHz 14/06/ :00 UT Delay (m s) Amplitude Figure Barker-13 sounding scattergram fo r Angle to Leicester path at 11:00 on 14/6/1997 a t 4.92 MHz. 1 Barker-13 sounding: MHz 14/06/ :00 UT < 0 R ece i ve d o o w erfd B r] Amplitude Figure Close up of Barker-13 sounding scattergram for Angle to Leicester path at 11:00 on 14/6/1997 at 4.92 MHz. 6-21

116 Stanag-4285 sounding: MHz 14/06/ :00 UT ^0.5 E < 0 R e c e ive d power (dbr] i: H J. / V W Amplitude Figure 6.24 STANAG-4285 sounding scattergram fo r Angle - Leicester path at 11:00 on 14/6/1997 at 4.92 MHz. Barker-13 sounding: MHz 14/06/ :30 UT R eceived power fdbrl 5 0 ' , l Amplitude Figure Close up o f Barker-13 sounding scattergram for Angle to Leicester path at 20:30 on 14/6/1997 at 4.92 MHz

117 Another data set was collected on 15/7/1997 and Figure 6.26 is a plot of the average measured modem throughput for a single data rate. This figure shows several periods where the modem decoded very little or no data at all. An analysis of the Barker-13 probing used on this day is presented in Figure 6.27 and shows the effective multipath spreads and Doppler spreads that were present. From this figure the Doppler and estimated delay spreads measured are both reasonably small across most of the day and should have caused few demodulation problems. Figure 6.28 is a plot of the number of interferers categorised as strong or weak that were detected on this day. This shows only a few incidents of strong interference being detected, however, those around the times of 08:30, 14:00,20:30 and 23:30 coincide with times of poor modem data throughput. The ionograms collected for this day indicate that propagation conditions for the frequency used were variable, often the only propagation mode detectable was probably due to sporadic E, this frequently resulted in a poor signal to noise ratio. The average data throughput at around 20:30 shows a significant drop compared with the preceding and following periods. The STANAG-4285 synchronisation sequence sounding channel evaluation technique was used to process the recorded signal at this time, the results are shown in Figure 6.29 where a large number of propagation modes can be seen to be detected. The delay spread at that time was at least 4.5 ms which is quite rare for this path. Interestingly this spread was not considered by the technique used to analyse the Barker-13 probe results as being a large effective delay spread. The modem will have failed to demodulate the received signal correctly due to the large number of modes present (at least five distinct modes). An ionogram recorded at the same time (Figure 6.30) shows 4 modes being present in the time range that can be displayed, the first being at a relative delay of 1.25 ms with the others following at 2, 2.75 and 3.75 ms. 6-23

118 150 STANAG bps 100 Mean Hour (UT) Figure Average throughput fo r STANAG bps on 15/7/1997 over the Angle to Leicester path » Hour (UT) Figure Effective delay spread and Doppler spread for 15/7/

119 STANAG bps GHD O' ' Weak interferers Strong interferers f2 <D S«1Z ? (r! o o q c p o m j > 9 1f U i * > i > 9-9 O 9 o c > <k > «.» 1 «y - ) Hour (UT) 20 Figure Num ber o f evident interferers fo r 15/7/1997. S tan ag-4285 sounding: MHz 15/07/ :30 UT < O' * R eceived power fdbr] Amplitude Figure STANAG-4285 synchronisation sequence sounding scattergram for Angle - Leicester path at 20:30 on 15/7/1997 at 4.92 MHz. 6-25

120 A n g le 1 5 /0 7 / : 3 0 U T ( ) MHz F igure Ionogram collected at 20:30 on 15/7/1997fo r Angle - Leicester path. Concluding remarks Ionospheric probing can reveal important channel information. This includes the number of propagation modes, the delay between each of them, any frequency shift between them. Therefore the delay and Doppler spreads can be calculated and used in order to aid the selection o f a suitable modulation scheme. Dedicated probing can be wasteful of transmit time. When the channel is used for sounding with a dedicated probe, the modem waveform cannot be used to transmit useful data. If dedicated probing is necessary, the timing of the probe sequence transmission should be matched with any higher level process that leads to a break in data transmissions. It is possible to make use of parts of some modem waveforms to gain channel evaluation information. Some data modulation schemes contain known signals which are intended to be used by a receiving modem to help train its equaliser. These signals are designed to estimate the impulse response of the channel and can therefore also be used for channel evaluation purposes if sampled for a short period. If the characterisation of all modulation schemes available for use is known, it is possible to switch between them to select the optimal scheme based on measured channel 6-26

121 evaluation information. The multi-dimensional characterisation method introduced above provides is a good way of describing a modem s BER behaviour under Doppler and delay spread conditions. Although the STANAG-4285 synchronisation sequence channel evaluation information appears limited in the Doppler range, such conditions are unlikely to be encountered where this modulation will be most used, i.e. over mid-latitude paths. A link maintenance system could make use of the parameters shown to be measurable in this chapter. It is recommended that such a system might be implemented in the following manner. After a data transmission period of reasonable length, say two minutes, the channel conditions should be assessed (the communications format described in STANAG sets a two minute maximum for transmissions). The format of the channel assessment will be dependent on the modem waveform used during the last data exchange. For example, if a waveform such as STANAG-4285, which contains a sequence readily useable as a channel probe, were to be used then channel parameters should be recorded periodically during the data transmission to ensure that an up to date set is available when the transmission ends. Alternatively, a probing phase may be entered whereby a waveform such as the Barker-13 probe should be sent from the transmitter. In this case, the probe should be analysed at the receiver and the channel parameters stored. Based on the results of either method of channel assessment, a decision should be made to select the most appropriate waveform to use for the next communications period. This decision should be signalled by the receiving system to the other end of the link using a 5-way handshake technique similar to that described in Chapter 5 (see Figure 5.8). The decision to change the waveform used for communications should be based on characterisations of modems using the technique described in this chapter, but might also be influenced by the needs of the communications system. In particular, the need to attain a greater communications speed might outweigh the need for robustness. Such a requirement will need to form an input to the system. 6-27

122 Chapter 7 - Channel probing by the addition of a pulse compression code to the modem waveform Channel evaluation by probing the ionosphere using coded pulse waveforms was demonstrated in the last chapter where the application of this to an automated HF linkmanagement system was discussed. It was concluded that performance benefits can be gained using such a technique. However, as also mentioned, using a dedicated probing waveform in place of the modem s modulation can be wasteful of transmit time and will lead to some loss in potential data throughput or an increase in system latency. Both of these effects are undesirable. Some modulation types, such as STANAG-4285, contain in-built sequences that can be used to some extent as channel evaluation probes. Where available, these can be used to aid link maintenance decision making without interrupting data transmissions. However, not all modulations contain in-built sequences that are suitable for use as probes, for example the traditional FSK modulation schemes. One reason for choosing the pulse compression waveforms in the previous chapter for channel evaluation was that they can yield information about the channel impulse response using only a moderate amount of transmit power. This achievement is due to spreading the energy of a transmitted pulse over many individual transmit symbols using bi-phase shift keying. This results in a processing gain when the received signal is passed through a matched filter. Additionally if several transmit codes are received and integrated over a short period of time, a further gain can be achieved without dilution of the channel parameter results. These two gains allow the probing signal and its information to be detected in the presence of a considerable amount of channel noise or interference. If a coded pulse sequence can stand out above the channel noise at the receiver, one with a larger gain figure may be detectable and useable at the receiver even in the presence of a data bearing modulation. Further to this, if the code was to be transmitted simultaneously with the data modulation signal, i.e. through the same transmitter, but at a significantly lower level, it could be useful as an add-on channel evaluation probe to any arbitrary modulation scheme without the need to break data transmissions. 7-1

123 To be useful as a probe, a pulse compression system needs to be able to measure delay and Doppler parameters, at least up to a threshold. In a realistic system, this threshold could be the point at which no modulation scheme available for use will work due to the channel conditions, but more likely will be somewhere prior to that point and will also be dependant on the purpose of the communications system. For example, a system that is to only ever operate in mid-latitude conditions is not likely to be affected by large Doppler spreads, but may well suffer from inter-symbol interference due to excessive delay spread. In another scenario, in this case where high-latitude paths are used, the more hostile ionospheric effects may often give rise to large Doppler spreads, and consequently the effects of these will dominate those of any delay spread. The parameters that are of greatest importance to a channel evaluation system which aims to maintain optimal data throughput under varying channel conditions are delay spread, Doppler spread and SNR. It is these three variables that have been used to characterise the operation of many modulation types as discussed in Chapter 6 because they have the greatest measurable effect on BER performance. However, emphasis must be placed on either delay or Doppler spread dependant on the likely path conditions, e.g. high or mid latitude. Different modulation types and their parameters such as the data rate require different levels of SNR to function acceptably. In Table 7.1 some of the common HF modulation types with illustrative data rates are tabulated along with the SNR required to achieve a BER of 1CT4 over a Gaussian noise channel (a BER of 10-4 is required for most frame based communications protocols used over HF to function well). These figures are useful when selecting a simultaneous probe waveform as they illustrate typical SNR levels that might be expected of a channel, and at which the probe should still be useful. 7-2

124 Modulation type SNR required for 10-4 (db) STANAG bps -1 STANAG bps 4.5 STANAG bps 11 MIL-STD B 3200 bps 7.5 MIL-STD B 9600 bps MIL-STD A FSK 600 bps 6.25 Table SNR requirements o f various modulation types to achieve a BER o f Iff4. Types o f probing sequences The length of a binary sequence that is used as an ionospheric probe has a bearing on how well that probe can be detected in the presence of noise or interference. As the length of a useful code increases, the processing gain also increases and this can allow the code to be detected in lower SNR conditions. To recap, the DAMSON system used a 13-bit Barker code as its probing sequence and this particular sequence yields a processing gain of just over 22 db. Other binary sequences exist with longer lengths and greater gains. One of these other sets of binary sequences that has good aperiodic autocorrelation functions was introduced in Chapter 2, these were discovered by Hu et al [1997] using neural network search techniques. These sequences are available up to at least 100 bits in length with lengths bits being presented in the paper. Such sequences could be used as probe waveforms if transmitted continuously, or with a gap between individual sequences if required to increase the delay range. The autocorrelation function of the 100 bit code discovered by this method is plotted in Figure 7.1. The peak to sidelobe ratio of this sequence is around 18.4 db. The Barker-13 code has a peak to sidelobe ratio around 22.3 db, this means that the results of using this code will be slightly less clearly defined if used for channel sounding. A further set of long binary codes was described by Gottesman et al [1992] in relation to low probability of intercept radar work where reducing peak transmit power was desirable. This family of codes are called modified Legendre sequences and were briefly introduced in Chapter 2. Modified Legendre sequences are a class of pseudonoise-like pulse 7-3

125 compression code pairs that exhibit Barker-like properties when cross correlated. Usefully for this application any sequence length L can be produced using this family of codes where L is a prime number of the form given in Equation 7.1 with m being an integer. L = 4m + \ (7.1) To make use of these codes a different approach is taken when detecting the code at the receiver. Instead of cross-correlating the known transmit sequence with the received signal to detect the pulse, as with the Barker sequences, with these sequences a modified version of the known transmit sequence is used in the cross correlation process. This second sequence is identical to the transmit sequence except that the first element has its sign reversed. The resultant cross-correlation response yields a single main peak with amplitude equal to L-2 and all other values flat at an amplitude of minus one. See Figure 7.2 for an example cross-correlation function of a modified Legendre sequence and it corresponding detection sequence for L=229. The peak to sidelobe ratio of this sequence is around 47 db c o 15 i Lag number F igure 7.1 Autocorrelation function o f neural netw ork search derived 100 bit sequence. 7-4

126 25 0 i r <D J TO J I I I I L Lag number Figure Cross-correlation fu n ctio n o f 229 bit m odified Legendre sequence. A 229 bit modified Legendre sequence for channel evaluation As these sequences offer good processing gain figures and good sidelobe performance, they are ideal candidates for a channel probing system to be used at the same time as a data modulation signal. However, having a probing system using a long sequence with a high gain will not be applicable to all channel conditions due to the effects of Doppler conditions. A series of experiments were conducted over the Angle - Leicester path to evaluate a 229-bit modified Legendre sequence as a potential channel probe. Limitations and theoretical performance of code A sequence length of 229 bits was chosen for use over reasonably benign mid-latitude skywave paths where the STANAG-4285 or MIL-STD B type modulations are most likely to be used. The SNR requirements for these modulations vary with data rate but are typically between -1 and 19 db as indicated in Table 7.1. The actual sequence length used here was one of several close possibilities, any of which could have been chosen. The length of these sequences determines their processing gains and their peak to sidelobe ratios. Based on a modem possibly requiring around 20 db SNR to function well, any simultaneous probe must be sent at 20 db below such a modem s waveform to 7-5

127 maintain the SNR. It is already known that the Barker-13 sequence with a gain of around 22 db yields good sounding results, therefore a sequence with a gain greater than 42 db should perform as well when transmitted at 20 db lower than the modem waveform, assuming the Barker-13 probes and the modem waveform are sent at the same level. The processing gain of a 229 bit Barker-like code is around 47 db, this should allow the probe to be transmitted at well below the data modulation level and yet still yield a large peak above the data signal when cross correlated. However, such a long code cannot work well in conditions of high Doppler spread. The code repetition rate is 10 codes per second at 2290 baud and as such can only measure Doppler spreads unambiguously up to +/-10 Hz. Additionally, due to the length of the code and hence the time taken to transmit it, Doppler effects can have a pronounced effect on the correlation result. For example, with a Doppler shift of 10 Hz, a phase shift of 180 will occur over half the code s length, thus a lot of the processing gain that the code was chosen for will be lost at the receiver. To determine the performance of this probe waveform under varying SNR conditions, a simulation was run where the probe was mixed with random noise of various levels and the resultant signal was cross-correlated with the detection sequence. The results in the format of peak to mean-minus-peak against SNR are plotted in Figure 7.3 below. This format was chosen to given a true measure of how distinct the correlation peak is above the noise level. From this figure, it can be seen that even at negative SNRs the peak should still be readily detectable. 7-6

128 0, , , , T SNR (db) Figure P eak to mean performance o f 229 bit modified Legendre sequence with varying SNR bit modified Legendre sequence probe format The 229-bit sequence to be transmitted was generated in a back to back format as in Figure 7.4, that is with one code following on immediately from another. The chip rate was 2290 chips per second. This non-standard rate was chosen to ensure that the phase was always at the same angle at the beginning of each code to allow a single code to be generated at the transmitter and repeated as required with no phase discontinuities. Any such discontinuities introduced at the transmitter would be likely to appear in the results as periodic noise. 229 bit code 229 bit code Figure Transmission form at fo r 229 bit modified Legendre sequence probe experiments. At the receiver, a sampling rate of four times greater than the chip rate, 9160 samples per second was employed to ensure the signal was faithfully recorded. Although a lower and more common-place sampling rate could have been chosen, having a sampling rate equal to four times the chip rate made the signal processing far simpler. Selecting the receiver system s sampling rate was achieved by means o f a simple configuration file entry. 7-7

129 To determine the channel conditions using this probe waveform, a technique was used to produce scattergram plots as first introduced in Chapter 5. An integration period of 2 seconds was employed in order to gain a good measure of the varying channel conditions. An example scattergram for this waveform under ideal channel conditions is reproduced here in Figure 7.5. C ode229 sounding: Sim ulated perfect channel "b R eceived power (dbr) Delay (m s) 5 r Amplitude Figure Scattergram p lo t fo r 229 bit m odified Legendre sequence probe under no-noise conditions. Experimental configuration To evaluate the performance of the 229-bit code over an HF path, a series of experiments were conducted over the Angle - Leicester path between September and December 1997 on a frequency around 5 MHz. The experimental equipment was arranged as described in Chapter 3 with the transmitter at Angle generating and sending the coded pulse waveform. The experiments had various formats designed to allow comparisons to be made between using a Barker-13 code, the STANAG-4285 synchronisation sequence, ionograms and the 229-bit probe for channel evaluation. The 229-bit probe was also transmitted simultaneously with a data modulation with varying relative amplitudes. 7-8

130 Results of probing with 229-bit modified Legendre sequence The transmission format chosen to allow the evaluation of the 229 bit sequence as a channel probe alone is illustrated in Figure 7.6, this shows a combination of waveforms which were transmitted serially for the given durations. Barker 13 probe (10 seconds) STANAG bps (90 seconds) 229 bit probe (90 seconds) STANAG bps plus 229 bit probe (90 seconds) Figure Transmission form at fo r evaluating 229 bit probe waveform on 14/10/1997. The first example of results from using this sequence as a probe in the above transmission format is taken from data collected around 08:00 hours on 14/10/1997 at a frequency around 5 MHz. Figure 7.7 shows the results from the Barker-13 sounding sent at the beginning of the transmission and this indicates one detectable mode of propagation and a high level of background noise. Additionally, in-channel interference is clearly seen in the frequency / delay / power pane of the figure. The 229-bit sequence was transmitted 110 seconds after the Barker-13 sequence, the resultant scattergram plot is shown in Figure 7.8. The single transmission mode, as in Figure 7.7, is again present but appears quite spread in the frequency domain. Figure 7.9 is a copy of the Barker-13 scattergram, but with the same frequency range on the y-axis as the 229-bit sequence scattergram to demonstrate that similar spreading o f the signal was detected by both methods. The inchannel interferer is also present in the 229-bit sequence scattergram, but only at a positive frequency offset. However, the noise floor appears much lower with this sounding, this is due to the much increased processing gain of the 229-bit sequence over the 13-bit sequence and because the amplitude scale is normalised to the peak value of the correlation function. To complete this data set, an ionogram collected around 08:00 is presented in Figure 7.10, this indicates a single propagation mode around the transmit frequency. 7-9

131 1 Barker-13 sounding: MHz 14/10/1997 8:00 UT Amplitude Figure Scattergram plot fo r Barker-13 probe at 08:00 hours 14/10/1997. C od e229 sounding: MHz 14/10/1997 8:00 UT R eceived power (dbr) Delay (m s) 5 N X Amplitude Figure Scattergram plot for 229 bit modified Legendre sequence probe at 08:00 hours 14/10/

132 Barker-13 sounding: MHz 14/10/1997 8:00 UT Received pow er (dbr) Pi a I n u ( Kv*i Amplitude Figure Close up o f scattergram plot fo r Barker-13 probe at 08:00 hours 14/10/ A ngle 1 4 /1 0 / :0 0 UT (096) M H 7 Figure Ionogram collected at 08:00 on 14/10/1997for Angle - Leicester path. 7-11

133 A second set of examples is taken from data collected later on the same day at 09:30 hours. The results of the Barker-13 probe are given in Figure 7.11, those of the 229-bit probe in Figure 7.12 and an ionogram for the same path at around the same time in Figure 7.13 All of these results indicate two modes of propagation present with around 0.65 ms of delay spread. These modes are only just distinguishable in the Barker-13 sounding results but are very clear in the results of the other two methods. Barker-13 sounding: MHz 14/10/1997 9:30 UT H Am plitude Figure Scattergram plot fo r Barker-13 probe at 09:30 hours 14/10/

134 Code229 sounding: MHz 14/10/1997 9:30 UT R eceived power (dbr) Delay (m s) Amplitude Figure 7.12 Scattergram p lo t fo r 229 bit modified Legendre sequence probe at 09:30 hours 14/10/1997. A ngle 1 4 /1 0 / :3 0 UT (114) MH? Figure Ionogram collected at 09:30 on 14/10/1997for Angle - Leicester path. 7-13

135 The effects of the 229 bit probe waveform on the performance of a data modulation In the previous section, the performance of the 229-bit sequence as a channel probe was demonstrated using a format where the probe waveform was transmitted by itself. The reasoning behind investigating this long sequence as a probe was to determine if such a waveform could be transmitted in conjunction with a data bearing modulation without adversely affecting the performance of that modulation. Therefore a series of experiments were performed to measure the effect of the probe waveform on the throughput rate of a typical HF data modulation scheme. The transmission format for these experiments which were conducted between 22/10/1997 and 26/10/1997 is shown in Figure Sixteen different waveform segments were sent during one transmit period and these consisted of a Barker-13 probe, a burst of STANAG-4285 modulation and nine bursts of STANAG-4285 modulation with the 229-bit sequence at varying levels in parallel. At the receiver, the data bursts were demodulated and analysed for block errors as first described in Chapter 4. Barker 13 probe (10 seconds) STANAG bps (30 seconds) STANAG bps plus 229 bit probe at 6 db (30 seconds) STANAG bps plus 229 bit probe at-7 db (30 seconds) STANAG bps plus 229 bit probe at-14 db (30 seconds) Figure Transmission form at fo r 22/10/ /10/1997. Figure 7.15 shows how a STANAG bps waveform performed over the course of 26/10/1997 in terms of data throughput. The same demodulation process was used to gather the throughput figures for the other data modulation bursts which were combined with the channel probe waveform. The general shape of the other plots are very similar, however, the throughput values vary and are summarised in Figure

136 150 STANAG bps 100 Mean #r m Cl 3Q. 46 Hour (UT) Figure D ata throughput f o r STANAG bps on 26/10/ No code bit code transmit level relative to data modulation Figure Data throughput fo r STANAG bps with various levels o f add-on code on 26/10/1997. It can be seen from Figure 7.16 that in general, increasing the signal level of the 229-bit sequence in relation to the data modulation has the effect of reducing the throughput of that 7-15

137 data modulation. This is due to the error rate increasing as the sounding waveform s level is increased. For this channel, it appears that using a probe level anywhere below 11 db down on the data signal has a negligible effect on the performance of the data modulation. However, if the channel s signal to noise ratio was different, then either greater or lesser probe levels could be tolerated. It is necessary to make a trade-off between code level and modem performance degradation. It would be sensible in any practical system to add the probing signal to the modem waveform at intervals for a short period of time in order to gather the channel parameters. There is little to gain from continuously probing if there are restrictions on how often modulation parameters can be changed, as there would be in any realistic communications system. As an example, STANAG-5066 allows a maximum transmission period of two minutes, and in Chapter 5 a five way handshake process that could be used to change parameters was described and this required around 40 seconds to complete. It might be reasonable to change parameters to prevent communications from being lost every two minutes, as allowed by STANAG-5066, but making changes to increase throughput must be undertaken with greater care to prevent transmission time being wasted on very short improvements in channel conditions. Excision o f add-on channel evaluation probe waveform The probe waveform described above for use as an add-on technique for channel evaluation is readily detectable and transmitted in such a manner that its starting phase is always the same. In other words each sequence sent is identical at the transmitter. Also, the application of this probe is to determine a measure of the channel s impulse response and therefore to identify how the channel has altered the probe between the transmitter and the receiver. With this information, it should be possible to subtract the received probe signal from the entire received signal to remove some or all of the probe energy that will appear to the data modem as noise. If that is possible, the effects of adding the probe waveform on the performance of the modem may be reduced or removed entirely. A technique has been developed to attempt to remove as much of the probe signal as possible at the receiver. This technique searches for the probe signal using the cross correlation method as described earlier in this chapter. The cross correlation process reveals peaks associated with sequences detected within the received data samples. A section of cross correlation results is plotted in Figure 7.17, this was taken at a time when 7-16

138 the 229-bit sequence was readily detectable. The peaks in this figure indicate the starting positions of individual probe sequences <o 5.1 O Lag Figure Peaks resultant fro m cross correlation detection o f 229 bit modified Legendre sequence. The correlation peaks can be used to estimate any frequency offset between the transmitter and receiver. In order to form this estimate, two correlation peaks need to be found which are separated by a distance d equal to that given by Equation 7.2 where s is the sampling rate, I is the sequence length and r is the baud rate the sequence was generated at. In the case of this 229 bit sequence, the distance is 916 samples. r d = sx (7.2) If two peaks are found separated by the above distance, the phase difference A</> between them can be calculated. The time difference between the peaks, A t, is known and hence a frequency offset (co) estimate can be made, see Equation 7.3. This frequency offset estimate is limited because the maximum phase difference that can be unambiguously measured is ± 7 t radians. 7-17

139 By measuring the amplitude of the peaks and any frequency offset across a reasonable length set of data samples, it is possible to produce an estimate of the signal that is required to cancel the received probe sequence. By adding this cancelling signal to the received samples at the correct positions (those relating to the detected sequences), some, if not all, of the energy of the probe sequence can be removed. The actual method used involved creating a cancelling signal for every two probe sequence periods based on an amplitude and frequency offset estimate for the same period to be cancelled. If estimates could not be made for a particular period, estimates for the previous period were used. This technique is currently very slow in operation and cannot be used for real time removal of the probe signal. With non-spread simulated signals, it is possible to remove a considerable mount of the probe signal energy. This has been quantified by observing the ESNIR from STANAG-4285 synchronisation sequence probe scattergrams, both before and after the excision technique was applied to the baseband samples. Some results are presented in Figure 7.18 in the format of SNR degradation due to adding the probe (i.e. before the excision process), SNR degradation after the excision process and the difference between the two (i.e. the SNR improvement that the excision process can yield). The results were gathered by measuring the ESNIR from the STANAG-4285 probe both with various levels of the 229 bit probe added (ranging from 0 db to 25 db down on the modem level), and for the special case where no code was added. It is evident that the greatest gain from the excision process was achieved for a 229 bit probe added at a level of around 5 db down on the modem signal. At probe levels below about db, the excision process actually reduced the SNR rather than improving it. The results for simulated signals were encouraging, however, for spread signals the ESNIR improvement obtained by using this excision process was much lower, typically around 0.5 db. 7-18

140 CO -O oc. z CO c<u cn cra O 5-20 SNR degradation before excision SNR degradation after excision SNR improvement due to excision Probe level relative to modem level (db) F igure Effect o f 229 bit probe on ESNIR before and after probe excision. Concluding remarks Three families of sequences that can be used as pulse compression codes for channel probing have been examined here: the Barker sequences, sequences found using a Neural netw ork search program and the modified Legendre sequences. The Barker-13 sequence has already been used as a dedicated probe in the DAMSON system, it is transmitted in place of a data modulation signal. The Barker family of sequences have a special property in that their sidelobes (the off perfect match response) are limited to being 0, 1 or -1 in value dependant on the sequence length. However, the Barker family only extends in length to 13 bits and this particular code has a maximum theoretical processing gain o f 22 db, a measured value will typically be lower. Longer sequences than offered by the Barker family of codes have been found using a Neural network search. These have quite good aperiodic autocorrelation functions (sidelobes) and have been found for lengths of up to 100 bits. Another set of sequences are known as modified Legendre sequences, these have two useful properties: they can be created for many lengths extending far beyond even those found by Neural network searches, and they have Barker-like sidelobe responses. 7-19

141 The performance of a long (229 bit) modified Legendre sequence in the presence of random noise has proven it to be readily detectable under all SNRs that modem modem waveforms can be expected to function under. Such a sequence could be transmitted simultaneously with a modem waveform to add the functionality of a probe waveform to a communications system. The effects of adding this sequence to a STANAG-4285 signal have been quantified by on-air tests that measured the data throughput achieved for various levels of probe signal. If the probe signal is more than about 11 db down on the modem waveform, little effect on the throughput was measured. It is not necessary to transmit the probe signal for the complete duration of the modem waveform as it is unlikely that modem parameters will be changed more often than around every two minutes. Therefore any degradation in modem performance could be limited to a very short duration. Although these long codes have many advantages, they are unfortunately not very tolerant to Doppler conditions as the ffequency-shifts-with-time affect the detection algorithm severely. The idea of removing the probe signal at the receiver by detecting it and adding a cancelling signal was investigated and proved to work very effectively on non-spread simulated signals. On spread signals, the process was far less effective and requires more research. Using an add-on channel evaluation system that does not require data communications to be halted could have enormous benefits to a communications system. As with the other probing methods described earlier, the modulation scheme parameters could be optimised for the current channel conditions by measuring the channel parameters. Uniquely, no additional time need be spent gathering these channel parameters as they can be measured whilst data is still being sent across the link no matter what modulation scheme is employed. 7-20

142 Chapter 8 - Conclusions and recommendations for further work Modem modem waveforms are now capable of achieving data rates of up to 12,800 bps within the bandwidth of a standard HF channel (3 khz), however the propagation characteristics are often such as to significantly reduce the achievable data throughput. In particular, these high speeds may only be achieved in conditions of small delay and / or Doppler dispersion and require a high signal to noise ratio. Currently available modems typically support several modulation standards, usually with a range of data rates. This enables the modem to operate under a variety of channel conditions due to the varying requirements of these schemes, however the selection of modem waveform is usually undertaken manually by system operators. In order to realise the best possible data throughput it is desirable that new systems should incorporate techniques to select automatically the most appropriate modem waveform to be employed at any given time in addition to the existing methods of frequency management. Consideration has been given to the problems associated with modem waveform management in this thesis. In particular: The variation in performance of differing modulation schemes; The channel requirements of various modulation schemes; Methods for selecting the suitable modulation scheme for the current channel. Before a waveform can be selected, some knowledge of the HF channel over which it is intended to be used needs to be obtained. This may be through the use of channel sounding systems (e.g. oblique ionosondes or fixed frequency channel probes) or prediction techniques (e.g. IONCAP, REC533). Observations of the performance of various modem waveforms over an HF path have shown that an ionosonde cannot provide enough information about the channel to allow accurate choices to be made when selecting a waveform to use. Furthermore, it is impractical to employ the use of an ionosonde over all communications paths due to the nature of the signals they use and the time taken to collect measurements. 8-1

143 The characterisation of the performance of modem waveforms under various levels of delay spread, Doppler spread and SNR has been demonstrated. Some of these results have been used in an automated system to examine the effects of changing modem data rates based on measured frame error rates. The results show that a significant improvement in data throughput can be achieved by using an adaptive data rate system when compared with a fixed data rate system. The automated data rate change system showed an improvement of over 9 times that which could be achieved using a fixed 75 bps data rate for the duration of a day. However, this system required a period of up to 40 seconds in order to switch between data rates and during this time no data transmissions can take place. Results from using a Barker-13 pulse compression waveform, as employed by DAMSON, for channel probing have been shown to provide accurate assessments of the current channel. These are relevant to the process of selecting a modem waveform for use. This type of probe has the requirement that data communications over a channel must be suspended whilst the channel parameters are measured. This is a limitation which will cause lower system throughput and increased message transmission latencies, both of which are undesirable. Some modem standards contain a probe signal as part of their normal transmission format. STANAG-4285 is one such example which uses an in-built probe to train a receiving modem s channel equaliser. This probe has been considered for use for channel evaluation and has been proven to be able to measure the channel parameters needed for waveform selection. The advantage of this technique is that data communications need not be halted in order to assess the channel. Transmitting a long, high gain, probe signal at a lower level than, and simultaneously with, a modem waveform has been investigated and demonstrated to be useful as a channel evaluation scheme. The concept of removing this add-on probe signal from the received signal prior to demodulation by a modem has been shown, however further development is required in order to achieve useful results when the received signal is subject to significant delay / Doppler spreading. Ideally, an excision algorithm would reduce the effective amplitude of the probe signal to levels which have an insignificant effect on the modem signal. 8-2

144 The effects of delay and Doppler spreads are of great importance in any modem waveform management system. Furthermore, the deleterious effects of different types of background interference on the various modem waveforms needs to be considered when choosing the most appropriate waveform to use. An effort must be made to identify the nature of any interference signals that are present on the channel of choice and to determine their likely effect on potential modem waveforms. The interference signals might be categorised by assessing their frequency spectra and their amplitude at the receiver. Similarly to the performance characterisation of modem waveforms for various channel parameters, the performance of these waveforms should be characterised for varying types of interference. In the architecture of a waveform management system the following points, as indicated in Table 8.1, are required. Columns A, B and C indicate respectively whether the requirements stated have been considered here, considered and require more work or whether the requirements have only been identified. Requirement A B C Characterisation of modem waveforms for delay & Doppler spread and SNR. Characterisation of modem waveforms in the presence of interference signals. Measurement of channel delay & Doppler spread. Measurement of channel SNR and fading characteristics. Characterisation of channel interference signals. Removal of add-on probe signal by excision. Use of propagation predictions to determine expected SNR and delay spread. Use geographical knowledge to determine likely Doppler effects. s / Automate selection of modulation type and data rate. Integrate with communications protocols. Consideration of the purpose of communications when selecting modulation type and data rate. Table The requirements o f waveform management system 8-3

145 Whilst several further developments and investigations have been identified, the results of this investigation could currently be employed in developing new systems to enhance the reliability and efficiency of HF data communications. For example, a STANAG-5066 / STANAG-4285 communications system could adapt the modem data rate based on the use of frame error rate statistics and on estimates of the channel delay spread, Doppler spread and SNR obtained using the in-built probe (training sequence) of the modem waveform. For systems that operate over time-varying channels and which may employ a wide range of modulation standards, long, modified Legendre probe waveforms may be transmitted periodically in conjunction with any modem waveform (when the waveform itself does not contain a suitable sequence) to determine the channel characteristics so that the data rate and / or the modem waveform can be adjusted such as to maximise the data throughput. An overview of such a system is presented here. An automated HF communications system The results presented in the previous chapters of this thesis might be incorporated into a fully automated communications system as shown schematically in Figure 8.1. In addition to the standard HF modem, drive, receiver and ALE components, the following are also required: a baseband filter, a probe generator, a channel monitor and a controlling PC. Probe generator ALE/ACS HF drive unit Controlling PC Multiwaveform modem Baseband filter HF receiver Channel monitor Figure Schematic o f an automated communications system. 8-4

146 At the most basic level, the modem, drive and receiver of this system can work as a traditional manually controlled station. With the addition of the ALE / ACS component and the controlling PC, a link can be established automatically with a remote station as required and on the best channel in terms of SNR. This link will allow a reasonably robust, starting waveform to be used, an example waveform is 300 bps STANAG The ACS component should monitor all available frequencies periodically to allow quick frequency changes to be made in the case of the current channel failing due to poor propagation or excessive interference. When the initial link is set-up, data can be transferred between stations. STANAG-5066 provides a standardised data link layer protocol for both broadcast (best attempt) and ARQ (reliable) modes of operation. It is likely that this will be appropriate for most HF communications needs owing to its flexibility and support for a diverse range of client applications. During data communications, link maintenance procedures need to be undertaken by the system to maintain and optimise the use of the channel. One method of doing this which is based on techniques presented in earlier chapters is described here. If a waveform such as STANAG-4285, which contains an in-built sequence suitable for use as a channel probe, is being used for communications, assessments of the current channel should be made periodically by means of sampling a short section of the baseband signal via the channel monitor component. These samples should be processed using a technique similar to that described in Chapter 6 to yield the channel parameters of delay spread, Doppler spread and SNR. Collecting these parameters often and taking an average over a period of time, say every few minutes, will result in a good indication of the channel s state. These results should be made available when a suitable transmission period comes to an end and changes in the modem waveform can be made. If in-built probing is not readily available from the modem waveform being used, an alternative strategy will need to be adopted in order to gather the channel parameters. The probe generator component in the schematic represents a device that can transmit a dedicated channel probe on the current frequency in use. A probe waveform such as the 229-bit modified Legendre sequence as described in Chapter 7 should be sent from the transmitting system and sampled at the receiving system using the channel monitor as described above. Such a probe can be transmitted whilst data communications are in 8-5